EP3742687A1 - Procédé et appareil de régulation de l'encombrement de réseau sur la base des gradients de vitesse de transmission - Google Patents

Procédé et appareil de régulation de l'encombrement de réseau sur la base des gradients de vitesse de transmission Download PDF

Info

Publication number
EP3742687A1
EP3742687A1 EP20184668.0A EP20184668A EP3742687A1 EP 3742687 A1 EP3742687 A1 EP 3742687A1 EP 20184668 A EP20184668 A EP 20184668A EP 3742687 A1 EP3742687 A1 EP 3742687A1
Authority
EP
European Patent Office
Prior art keywords
congestion
trend
transmission rate
flight size
tcp
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
EP20184668.0A
Other languages
German (de)
English (en)
Other versions
EP3742687B1 (fr
Inventor
José LOPEZ SERRANO
Luis PIÑEIRO BLANCA
Guillermo Lopez Serrano
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Bequant SL
Original Assignee
Bequant SL
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Family has litigation
First worldwide family litigation filed litigation Critical https://patents.darts-ip.com/?family=50685847&utm_source=google_patent&utm_medium=platform_link&utm_campaign=public_patent_search&patent=EP3742687(A1) "Global patent litigation dataset” by Darts-ip is licensed under a Creative Commons Attribution 4.0 International License.
Application filed by Bequant SL filed Critical Bequant SL
Publication of EP3742687A1 publication Critical patent/EP3742687A1/fr
Application granted granted Critical
Publication of EP3742687B1 publication Critical patent/EP3742687B1/fr
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L47/00—Traffic control in data switching networks
    • H04L47/10—Flow control; Congestion control
    • H04L47/11—Identifying congestion
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L47/00—Traffic control in data switching networks
    • H04L47/10—Flow control; Congestion control
    • H04L47/12—Avoiding congestion; Recovering from congestion
    • H04L47/122—Avoiding congestion; Recovering from congestion by diverting traffic away from congested entities
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L47/00—Traffic control in data switching networks
    • H04L47/10—Flow control; Congestion control
    • H04L47/12—Avoiding congestion; Recovering from congestion
    • H04L47/127—Avoiding congestion; Recovering from congestion by using congestion prediction
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L47/00—Traffic control in data switching networks
    • H04L47/10—Flow control; Congestion control
    • H04L47/15—Flow control; Congestion control in relation to multipoint traffic
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L47/00—Traffic control in data switching networks
    • H04L47/10—Flow control; Congestion control
    • H04L47/16—Flow control; Congestion control in connection oriented networks, e.g. frame relay
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L47/00—Traffic control in data switching networks
    • H04L47/10—Flow control; Congestion control
    • H04L47/17—Interaction among intermediate nodes, e.g. hop by hop
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L47/00—Traffic control in data switching networks
    • H04L47/10—Flow control; Congestion control
    • H04L47/18—End to end
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L47/00—Traffic control in data switching networks
    • H04L47/10—Flow control; Congestion control
    • H04L47/19—Flow control; Congestion control at layers above the network layer
    • H04L47/193—Flow control; Congestion control at layers above the network layer at the transport layer, e.g. TCP related
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W8/00—Network data management
    • H04W8/02—Processing of mobility data, e.g. registration information at HLR [Home Location Register] or VLR [Visitor Location Register]; Transfer of mobility data, e.g. between HLR, VLR or external networks
    • H04W8/04—Registration at HLR or HSS [Home Subscriber Server]

Definitions

  • the present invention relates to the field of network communications and, more particularly, to a method and an apparatus for congestion control in network communications.
  • Fig. 1A shows a data transmitting node 110 sending a data flow 150 to a data receiving node 120 through a communications network 140, which includes a collection of interconnected network nodes 141, 142, 143, and possibly others.
  • the data flow 150 is relayed in this case by network nodes 141 and 142.
  • Fig. 1A further shows another data transmitting node 111 sending a data flow 151 to a data receiving node 121 through the same network 140. However, the data flow 151 is relayed in this case by network nodes 141 and 143.
  • the aggregated sent rate of data flows 150 and 151 is larger than the relaying capacity of network node 141, in which they coincide, then network node 141 will become congested and both data flows 150 and 151 will experience a congestion situation. If a network node (such as 141) has some input buffering capability, the congestion situation will cause its buffering capacity to be progressively used until it may eventually fill up. Once filled up, if the congestion situation persists, part of data flows 150 and 151 will be discarded and thus lost from the point of view of the data receiving nodes 120 and 121. Congestion control mechanisms are used to avoid, mitigate and handle congestion situations.
  • Congestion control mechanisms can be part of the functionality of any layer, typically the link layer, the network layer, the transport layer or the application layer, and can reside in the end nodes (such as 110 or 111) or in the intermediate nodes (such as 141, 142 or 143).
  • Transfer Control Protocol uses sophisticated congestion control mechanisms. Many of the technical advances in the field of congestion control have taken place within the development of TCP. Thus, in the following, congestion control mechanisms in TCP will be described. However, many of the technical developments coming from TCP have later been adopted by other protocols, such as Stream Control Transmission Protocol (SCTP).
  • SCTP Stream Control Transmission Protocol
  • Transport Control Protocol is a widely used in data communication networks.
  • the TCP specified in J. Postel, "IETF RFC 793: Transmission control protocol," 1981 , and incorporated herein by reference, provides reliable data transmission between two endpoints. Endpoints are commonly referred to as "hosts” in literature concerning TCP technology.
  • the term “reliable data transmission” refers to the fact that the TCP provides an adaptive repeat request (ARQ) mechanism enabling an acknowledged data transmission.
  • ARQ adaptive repeat request
  • a data transmitting node 110 transmits a data segment 101 (payload of TCP datagram) through a network 140, of which a data receiving node 120 checks the correct arrival.
  • the data receiving node 120 then sends back to the data transmitting node 110, through a network 140, an acknowledgement 102 positively acknowledging the correct reception of the data.
  • the data transmitting node 110 may retransmit the data.
  • the acknowledgements are also transmitted in TCP segments and may be cumulative, i.e. the acknowledgement of a TCP segment implies the acknowledgement of all prior consecutive TCP segments.
  • Data packets may get lost, which means that they do not arrive within a predetermined time window (i.e. time period) at the data receiving node.
  • data packets may experience some transmission errors, which may be detected at the data receiving node by standard means including error detection/correction codes such as cyclic redundancy check (CRC) or others.
  • error detection/correction codes such as cyclic redundancy check (CRC) or others.
  • the delay or transmission errors may be caused by increased load within the network and/or by worsening of the channel conditions.
  • the acknowledgement mechanism provided by the TCP enables recovering the packet loss and corrupted data by means of retransmissions.
  • TCP provides some congestion control mechanisms and strategies, which may be implemented on the data transmitting node 110 and/or data receiving node 120.
  • Congestion Window denotes the maximum amount of consecutive data a TCP host can send beyond the latest acknowledged sequence number, as calculated locally by the sending host, not having taken into account the window advertised by the receiving host.
  • Receive Window is the window size advertised by the receiver to the sender in the acknowledgement messages it sends back. This window size specifies the maximum amount of consecutive data the receiver is ready to accept beyond the latest acknowledged sequence number. This is the mechanism used by TCP to implement flow control, i.e. to prevent a fast sender from overwhelming a slow receiver.
  • Send Window is the lowest of cwnd and rwnd.
  • Standard TCP congestion control is based on mechanisms for the sender, and thus focuses on setting an appropriate cwnd value, assuming that cwnd determines the Send Window.
  • Slow Start is a TCP congestion control state.
  • the TCP congestion control algorithm increases the cwnd in an exponential fashion, where cwnd is increased by about one segment size every time an acknowledgement is received.
  • a TCP sending host in a TCP Connection is said to be in Slow Start when it is using this manner of increasing the cwnd.
  • Congestion Avoidance is a TCP congestion control state.
  • the TCP congestion control algorithm increases the cwnd in a slower fashion than in Slow Start.
  • the cwnd grows by 1/cwnd bytes (the inverse of cwnd) for every byte acknowledged, i.e. cwnd will grow by the size of one segment after acknowledging segments that add up cwnd bytes in size.
  • Different TCP variants have different congestion avoidance algorithms.
  • a TCP sending host in a TCP Connection is said to be in Congestion Avoidance when it is using this manner of increasing the cwnd.
  • Slow Start Threshold defines the transition between Slow Start and Congestion Avoidance control states: it is the Congestion Window value below which Slow Start is used and beyond which the Congestion Window is increased according with Congestion Avoidance, Flight size is the amount of data, usually measured in bytes, transmitted by the sender and not yet acknowledged.
  • Buffer-bloat is a term for an excessive use of buffers in the transmission path by a connection using a larger congestion window than what would be required to overcome the intrinsic delay and bandwidth adaptation buffering in the transmission path.
  • TCP terminology described above may also be used in other protocols which may use similar concepts as those used for TCP congestion control.
  • TCP specifications such as IETF RFC 5681 cited above; IETF RFC 6582 "The NewReno modification to TCP's fast recovery algorithm," from 2012 ; IETF RFC 2018: TCP selective acknowledgment options” from Oct. 1996 ; and IETF RFC 6675: "A Conservative Loss Recovery Algorithm Based on Selective Acknowledgment (SACK) for TCP” from 2012 (all incorporated herein by reference) include congestion control algorithms for determining the most appropriate data sending rate for the hosts and data recovery mechanisms to allow the efficient retransmission of data lost due to congestion or to other causes.
  • SACK Selective Acknowledgment
  • the TCP congestion control approach is derived from the so-called “Reno” congestion control mechanism (described, for instance, in V. Jacobson, "Congestion avoidance and control,” in ACM SIGCOMM Computer Communication Review, 1988 ), with several additions to improve the recovery from packet transmission losses, and is based on the following principles:
  • This standard TCP (sometimes called Reno or NewReno), has three limitations: (1) low performance in high-speed/long-delay networks, because the small rate of growth of the Congestion Window in Congestion Avoidance takes a very long time to achieve the large sizes of Congestion Window required by those networks; (2) excessive use of network buffers (buffer-bloat), increasing the delay experienced by the communicating hosts, due to the loss-based mechanism to detect congestion, which increases the Congestion Window up to the point where the buffers in the network nodes in the path are full and therefore sent packets get dropped, and (3) competition against concurrent TCP "greedy" flows, i.e. TCP flows that raise their Congestion Window in a more aggressive way than this standard TCP when they are sharing a bandwidth bottleneck, which will take most of the available bandwidth for themselves and starve Reno TCP flows.
  • TCP linear Congestion Window growth in Congestion Avoidance
  • BIC-TCP BIC-TCP
  • H-TCP CUBIC
  • TCP-Hybla TCP variants
  • TCP variants that detect network congestion by analyzing the connection's end-to-end delay through the measured Round-Trip Time (RTT), the time from sending a segment to receiving an acknowledgement for it.
  • RTT Round-Trip Time
  • Some of such variants are TCP Vegas, TCP Vegas-A, TCP New Vegas, and FAST-TCP.
  • the RTT measurements are sometimes translated into sending rate estimates, or into estimates of segments queued (buffered) in the transmission path, but in reality the independent variable they use to make decisions are RTT measurements.
  • TCP-DUAL adds to a loss-based Reno congestion control a congestion detection algorithm based on RTT measurements, which triggers a multiplicative decrease in Congestion Window.
  • This approach can solve the buffer-bloat problems, but because of the RTT-based delay detection, like the above mentioned pure delay-based variants, it cannot compete against loss-based TCP variants.
  • TCP variants like Compound TCP, TCP Libra, TCP Africa, TCP Veno, YeAH-TCP and TCP Illinois, also use mixed models with loss-based and delay-based congestion control.
  • the delay-based congestion detection is used to modulate the aggressiveness of Congestion Window growth, allowing more aggressive growth when no congestion is detected, which usually solves the performance problems associated to Reno/NewReno in high-speed/low-delay networks.
  • RTT-based metric estimates that there is congestion
  • the Congestion Window will still keep growing, albeit more slowly, until packet losses arise, so buffer-bloat problems will persist, even when not competing with other data flows.
  • TCP Vegas+ is a mixed model using TCP Vegas by default, but switching to NewReno if a competing flow is detected. This should avoid buffer-bloat problems when there is no competition from other flows, but several unsolved Vegas problems remain, for example, low performance in high-speed/long-delay networks.
  • a TCP variant called Tri-S is an early TCP variant with rate-based congestion detection, using the time evolution of the transmission rate.
  • Tri-S reduces the Congestion Window upon congestion detection, it cannot compete against a TCP flow with loss-based congestion control, just like delay-based TCP variants.
  • TCP variants denoted as TCP-Westwood and TCP-Westwood+ introduced explicit bandwidth estimations into their congestion control mechanisms, based on a complex measurement of inter-acknowledgement timing or simply the rate of received acknowledgements. With appropriate filtering, those estimates at the time a congestion-induced loss happens are taken as the available bandwidth for the TCP connection. This bandwidth estimation, together with the minimum RTT measured, is then taken to determine the optimum Congestion Window. From there on, a Reno-like Congestion Avoidance mode will eventually take the Congestion Window size to congestion and packet losses (thus causing buffer-bloat), at which point a new optimum Congestion Window will be calculated based on a new bandwidth estimation.
  • TCP-Westwood aimed at high-speed/long-delay networks (such as LogWestwood+, TCPW-A, TCP-AR and TCP Fusion) with more aggressive window growth, which adapt better to changing network bandwidth, but which still suffer from buffer-bloat and may overwhelm concurrent, less-aggressive Reno flows.
  • the transition from the Slow Start to Congestion Avoidance can be very important, especially in small downloads that spend a significant part of their existence in Slow Start. Detecting congestion in Slow Start only with losses may lead to severe buffer-bloat problems and packet losses, since congestion will be reached while the sending window is growing exponentially.
  • TCP-Vegas proposes a modified Slow Start which in fact causes a premature Congestion Avoidance due to the burstiness of the Slow Start traffic.
  • “Limited Slow start” is an experimental IETF RFC that relies on an arbitrary constant to determine the transition point.
  • Adaptive Start being a part of the TCPW-A variant, uses the estimated bandwidth to derive the Slow Start Threshold parameter, and thus strongly depending on the quality of that estimate.
  • Hybrid Start which is nowadays used by default in most versions of the wide-spread Linux operating system, uses two heuristic algorithms based on RTT measurements and inter-acknowledgement delays. This works well except when competing against concurrent TCP flows with congestion, because the concurrent traffic will increase the RTT from the beginning, and so the switch from Slow Start to Congestion Avoidance will occur too early, which will cause a decreased throughput.
  • TCPW-A which has a mechanism to increase the Slow Start threshold parameter if it is estimated that it would result in higher bandwidth, but it is conditioned by another mechanism to detect that there is no other TCP flow competing.
  • TCP congestion control approaches mentioned so far rely on functionality implemented in end hosts, primarily on the sender side. However, some approaches rely on functionality in intermediate network nodes, like routers and switches, which could alert the endpoint about impending congestion (e.g. TCP ECN), or drop packets before congestion occurs (e.g. Random Early Detection queue management algorithms).
  • TCP ECN Transmission Control Protocol
  • Random Early Detection queue management algorithms e.g. Random Early Detection queue management algorithms.
  • CoDel K. Nichols, V. Jacobson, "Controlling queue delay", Communications of the ACM, vol. 55, no. 7, pp. 42-50, 2012
  • CoDel K. Nichols, V. Jacobson, "Controlling queue delay", Communications of the ACM, vol. 55, no. 7, pp. 42-50, 2012
  • buffer-bloat remains an important problem in TCP communications, causing unnecessary delays and resource usage, because the most widely used TCP variants utilize loss-based congestion detection, which at the same time prevents the use of delay-based variants that could mitigate the buffer-bloat problem.
  • TCP-variants that use bandwidth-based and rate-based algorithms in determining the Congestion Window still use loss-based congestion detection, so buffer-bloat problems remain.
  • rate-based congestion detection is used to reduce the congestion window, measurement filtering and robust statistical methods are missing, so that the variability of measurements is not correctly addressed, and congestion is not well detected.
  • the particular approach of the present invention is to detect the congestion on the basis of the trends of both the flight size and the transmission rate in order to adapt the congestion window in accordance with the detection result.
  • Such approach has the advantage of reliable congestion detection and the possibility of distinguishing between congestion with unfair competition and congestion without unfair competition, situations which benefit from different treatment.
  • the competition from a concurrent data flow is considered unfair when it steadily takes up more and more of a congestion-limited bandwidth.
  • the filtering applied in connection with determining the trend of transmission rate and/or flight size ensures that the decision on presence of the congestion is performed stably with respect to the temporal variations of the measurements.
  • a method for congestion control in a data communication protocol employing acknowledged communication in which a transmitting node transmits data to a receiving node and the receiving node acknowledges receipt of the data the method to be performed by the transmitting node and comprising: measuring a flight size indicating an amount of data sent by the transmitting node and not yet acknowledged by the receiving node; measuring a transmission rate, the transmission rate corresponding to a same time instant as the flight size; determining a trend of the flight size; determining a trend of the transmission rate, wherein the trend is derived from a transmission rate gradient calculation, in which either the transmission rate measurements or the transmission rate gradient calculations or both, are filtered to reduce their temporal variability; detecting whether or not there is a congestion according to the determined trend of the transmission rate and the trend of the flight size; upon positive detection of the congestion, modifying the flight size; and transmitting data while respecting a maximum amount of unacknowledged data which the transmitting mode may transmit.
  • an apparatus for controlling, in a transmitting node, congestion in a data communication protocol employing acknowledged communication in which the transmitting node transmits data to a receiving node and the receiving node acknowledges receipt of the data
  • the apparatus comprising: a flight size measuring unit for measuring a flight size indicating an amount of data sent by the transmitting node and not yet acknowledged by the receiving node; a transmission rate measuring unit for measuring a transmission rate, the transmission rate corresponding to a same time instant as the flight size; a flight size trend calculation unit for determining a trend of the flight size; a transmission rate trend calculation unit for determining a trend of the transmission rate, wherein the trend is derived from a transmission rate gradient calculation, in which either the transmission rate measurements or the transmission rate gradient calculations or both, are filtered to reduce their temporal variability; a congestion detection unit for detecting whether or not there is congestion according to the determined trend of the transmission rate and the trend of the flight size; a congestion handling unit for, upon positive detection of the congestion by the congestion
  • Efficient packet-switched communications require some buffering in network nodes to allow switching, to adapt different link rates and to absorb temporary transmission bursts.
  • protocols using loss-based congestion control tend to use all the available buffers in the transmission path, beyond what is necessary for efficient communications, wasting memory resources and causing unnecessary delays, and resulting in buffer-bloat.
  • Buffer-bloat may in fact be unavoidable in certain circumstances, as when competing against other communication data flows under congestion, since in those cases the flows that try to limit buffer-bloat will be overwhelmed by those that do not. However, buffer-bloat could be avoided when not competing against other data flows for congested resources (a fairly common situation).
  • the present invention provides methods and apparatuses for network congestion control that tackle the buffer-bloat problem when it can be avoided without penalizing throughput. It adds as a further advantageous feature an estimation of transmission rate and flight size trends, together with a congestion detection mechanism, that is robust against observed variability. Particular embodiments of the present invention specify further approaches to detect congestion using the measured trends of flight size and transmission rate, which enables distinguishing between different congestion situations. Thus, congestion situations with competing data flows with different perceived aggressiveness may be treated differently.
  • the present invention can readily be employed within the TCP protocol and some of the embodiments also envisage incorporation into the TCP congestion handling algorithms. Nevertheless, the present invention is in no way limited to employment in the TCP and is applicable to any communication protocols that provide acknowledged transmission and make use of congestion control, such as Stream Control Transmission Protocol (SCTP), Datagram Congestion Control Protocol (DCCP) or others. Moreover, the present invention is applicable at any protocol layer and not limited to the transport layer.
  • SCTP Stream Control Transmission Protocol
  • DCCP Datagram Congestion Control Protocol
  • the present invention is applicable at any protocol layer and not limited to the transport layer.
  • both the flight size and the transmission rate are measured and used to determine their trends, with which it is determined whether or not there is a congestion situation. Based on the congestion detected, the strategy for modifying the transmission window size is selected.
  • the trend of the transmission rate is derived from its gradient.
  • the gradient may be any metric that captures the trend. It can be, for example, the difference between two values measured at two points in time, or the difference between estimates (like the mean, the median, the exponentially weighted mean of a collection of measurements) at two points in time, an estimate of the slope of a curve fitted to a sequence of measurements, the (weighted) average of several consecutive gradient calculations or the metric from a trend estimation statistical test (like the well-known Jonckheere's trend test).
  • the gradient can be any metric that can indicate if a sequence of values has an upward trend, a stable trend or a downward trend.
  • this metric will be measured repeatedly, in real-time, starting from measurements which may potentially include very significant amounts of noise, especially in the case of the transmission rate measurements. Therefore, either the input transmission rate measurements, or the transmission rate gradient (trend metric) or both, are filtered to reduce the effects of noise, which could otherwise hinder the congestion detection.
  • the trend of the flight size may be determined as any measure representing the evolution of the flight size or filtered flight size in time.
  • a gradient as described above, may be used.
  • the present invention provides the possibility of competing with a loss-based TCP flow over a bottleneck, in which case it will tend to match the aggressiveness of competing flows, without overwhelming them.
  • the approach is to measure, from the data sender's point of view, the flight size (i.e. the window being used) and the transmission rate attained with that flight size, then to determine their trends (time evolution) with statistically robust techniques, and finally to act upon the congestion window when it detects that the connection is in one of predefined congestion conditions.
  • the invention is applicable to TCP senders (hosts), to which it adds the capability of detecting the aforementioned conditions and, upon the event of detecting one of those conditions, it appropriately changes the congestion window or the algorithm that controls the congestion window growth.
  • the present invention employs statistically robust techniques and filtering in the estimation of transmission rate trends.
  • the five specific conditions that may be detected and acted upon are: (1) congestion without unfair competition in Slow Start, (2) congestion without unfair competition in Congestion Avoidance, (3) congestion with unfair competition in Congestion Avoidance, (4) congestion with unfair competition after a congestion window reduction, and (5) congestion with unfair competition after a transition from Slow Start to Congestion Avoidance. It is noted that the present invention is not limited to evaluating all of the above conditions and to handling them appropriately. Rather, any of the above conditions may be distinguished by estimating the trend of flight size and transmission rate according to the invention. Thus, each of the above five conditions, their detection and handling constitute a separate embodiment. Moreover, these particular embodiments may be combined with each other as is clear to those skilled in the art.
  • the invention is not limited to TCP and may be used for corresponding states (as defined in the above Background section or the corresponding TCP specifications) in other congestion control mechanism or apparatus.
  • the congestion state may be generally detected without distinguishing between congestion states such as Slow Start or Congestion Avoidance. Nevertheless, the distinction of the states further improves finer adaptation of the congestion control mechanism to the network conditions.
  • unfair competition is relative to the sending host that is performing the congestion detection and to its current state, i.e., it indicates that the competing data flow is more aggressive than the sending host in the current state.
  • more aggressive means that the competing data flow increases its congestion window (or its transmission rate) at a faster rate than the sending host.
  • a sending host may detect condition (1), i.e. congestion without unfair competition in Slow Start, when it is competing against another flow which is also in the TCP slow start state. Such competition will possibly not be considered as unfair, because it is just as aggressive as the sending host. However, if the sending host switches to Congestion Avoidance, it would start seeing the competing flow which remains in Slow Start as "unfair competition", since such flow would then appear more aggressive to the sending host.
  • Congestion is also relative to the sending host, in the sense that it refers to the symptoms of congestion detected by the sending host according to certain data transmission parameters that are measured and evaluated. Congestion is thus considered to be positively detected if certain predetermined condition(s) based on the measured parameters is/are fulfilled.
  • the measured parameters are flight size and transmission rate, together with their evolution evaluated over time (i.e. the trends) of those two parameters.
  • An embodiment of the present invention can (positively) detect condition (2), i.e. congestion without unfair competition in Congestion Avoidance state, by detecting a simultaneous increase in flight size and a stabilization in transmission rate. If condition (2) is positively detected, it is treated by reducing the congestion window. In the absence of Receiver Window or host buffer limits, the congestion window will govern the Send Window, which may then govern the flight size. Reducing the congestion window will therefore tend to reduce the flight size, which in turn will tend to mitigate buffer-bloat,
  • an embodiment of the present invention can (positively) detect condition (1), i.e. congestion without unfair competition in Slow Start, by detecting a simultaneous increase in flight size and a stabilization in transmission rate, which is treated by changing from the Slow Start to the Congestion Avoidance state. This will tend to make the flight size growth slower and, therefore, it tends to mitigate buffer-bloat.
  • condition (1) i.e. congestion without unfair competition in Slow Start
  • a simultaneous increase in flight size and a stabilization in transmission rate which is treated by changing from the Slow Start to the Congestion Avoidance state. This will tend to make the flight size growth slower and, therefore, it tends to mitigate buffer-bloat.
  • the positive detection of the congestion without unfair competition may still be performed by detecting the increase of the flight size and, at the same time, substantially unchanged transmission rate.
  • the handling may consist in reducing the congestion window or reducing the speed of congestion window growth.
  • an embodiment of the present invention can detect condition (3), i.e. a congestion with unfair competition situation in Congestion Avoidance, in which there is congestion and there is another data flow competing for capacity bottleneck resources in a more aggressive way (such as by having its congestion window growing faster).
  • condition (3) i.e. a congestion with unfair competition situation in Congestion Avoidance, in which there is congestion and there is another data flow competing for capacity bottleneck resources in a more aggressive way (such as by having its congestion window growing faster).
  • Such congestion with unfair competition may be (positively) detected, in Congestion Avoidance (CA) state, by detecting a simultaneous increase in flight size and decrease in transmission rate.
  • Positive detection of unfair competition in CA can be beneficially treated by either more aggressively increasing the Congestion Window or by switching from Congestion Avoidance to Slow Start.
  • the rate-based congestion detection can detect congestion when it occurs, normally much earlier than when congestion-induced losses occur, and buffer-bloat can be avoided by reducing the congestion window or transitioning from Slow Start to Congestion Avoidance at that point.
  • One of the advantages of detecting congestion based on flight size and transmission rate evolution in time is that it allows distinguishing between congestion with unfair competition and congestion without unfair competition, especially in Congestion Avoidance. Accordingly, the action to take in each case may be quite different, which improves the effect of congestion handling.
  • congestion is detected by a host (by the methods or elements of the present invention, or by other algorithms, like delay-based mechanisms or by the receipt of three duplicate acknowledgements), it may happen that in addition to the congestion situation, one or more data flows are competing for the available bandwidth, with a similarly aggressive window growth rate. Up to that point, the similarly aggressive flows could have been sharing the congested bottleneck with a roughly constant transmission rate, while their respective flight sizes kept rising in parallel (using the available network buffering).
  • the congestion window is reduced or the connection in Slow Start is switched to Congestion Avoidance, the now less aggressive flow will lose some transmission rate, which will be taken up by its competing data flows.
  • This situation is therefore called “unfair competition” in this description, since from the point of view of the data flow reducing its aggressiveness it is unfair.
  • This congestion with an unfair competition situation can be detected by the sending side a posteriori, evaluating the immediate results of lowering the congestion window or switching to Congestion Avoidance.
  • An embodiment of the invention can positively detect "congestion with unfair competition" after a congestion window reduction (i.e. condition (4) mentioned above) when, a short time after reducing the congestion window (following the positive detection of congestion), the flight size is not smaller than the amount of reduction in the congestion window, while at the same time, the transmission rate is lower.
  • the congestion window reduction may beneficially be undone, i.e. reversed.
  • an embodiment of the present invention can positively detect "congestion with unfair competition" after a transition from Slow Start to Congestion Avoidance (corresponding to condition (5) mentioned above) when, a short time after switching from Slow Start to Congestion Avoidance (following the detection of congestion), the flight size is not smaller and there is a reduction in transmission rate after the transition from Slow Start to Congestion Avoidance. If this condition (5) is detected, the connection may beneficially be switched back to Slow Start. In this way, the detection to validate the transition from Slow Start to Congestion Avoidance is able to determine that there is a reduction in transmission rate if there is a competing loss-based flow in Slow Start, in which case the flow would go back to Slow Start.
  • the congestion window can be set to the value stored in that variable.
  • Incorporating the invention to an existing TCP implementation, with its capability to detect the five above described conditions and their associated actions, may produce a TCP implementation that solves the buffer-bloat problem.
  • some TCP implementations may benefit from incorporating just a subset of those five condition detection capabilities (and associated actions). The same applies for congestion control mechanisms of other protocols which may be also implemented on other layers than the transport layer.
  • a TCP proxy node 130 is a network node that acts as an intermediate node between two end hosts 160 and 170 communicating via TCP.
  • the TCP proxy 130 terminates the TCP connection towards the two end hosts, in a transparent way with respect to the end hosts 160, 170.
  • This kind of proxy node 130 can relay the information sent from one end host to another end host (105 and 106 in Fig. 1C ) transparently or may try to add value by modifying it in some way.
  • a proxy node 130 provides may be beneficial since the configuration or implementation of the end hosts 160 and 170 does not need to be changed in order to benefit from the advantages of the present invention.
  • TCP includes many aspects beyond congestion control. Those other aspects do not need to be changed by the present invention. This invention may thus be easily adapted to an existing TCP implementation with loss-based congestion detection, adding certain steps at a point in the processing that will be specified below.
  • Figure 2 is a flow chart illustrating processing logic performed upon reception 210 of new acknowledgement segment (new ACK) according to an embodiment of the present invention.
  • a "new ACK” refers to an ACK segment which cumulatively acknowledges TCP user data previously sent by the destination host of this new ACK segment for the first time (i.e. that said data has not been cumulatively acknowledged before). It is noted that even when it is beneficial to perform the method after each reception of a new ACK, the present invention is not limited to such implementation. In general, the invention may be implemented performing the method less frequently than for each new ACK reception. It could even be performed not when new a new ACK is received, but periodically according to an internal clock, which would still allow it to estimate transmission rates.
  • TCP protocol header further includes, inter alia, flags such as ACK (indicating significance of the acknowledgement field), FIN (no more data from sender), SYN (synchronize sequence number, only set for the first packet), RST (reset of the connection) and others (described in detail in the above cited RFC 793).
  • flags such as ACK (indicating significance of the acknowledgement field), FIN (no more data from sender), SYN (synchronize sequence number, only set for the first packet), RST (reset of the connection) and others (described in detail in the above cited RFC 793).
  • ACK indicating significance of the acknowledgement field
  • FIN no more data from sender
  • SYN synchronize sequence number, only set for the first packet
  • RST reset of the connection
  • Step 220 refers to processing for a new ACK segment that a TCP protocol (or another protocol) may perform, as described before, up to the estimation of RTT.
  • step 220 may include reading the segment from a network interface, decoding its data-layer and network-layer headers, verifying whether the segment is a correctly formed TCP segment, etc..
  • SND.UNA in step 220 refers to a TCP connection state parameter, namely, to the sequence number of the lowest byte not yet cumulatively acknowledged, which is updated with the acknowledgement sequence in the new ACK message received.
  • the parameter cwnd denotes the congestion window size in bytes.
  • a TCP host usually calculates (step 230) the RTT.
  • RTT is the Round-Trip Time: the time since a certain TCP segment is sent until the first acknowledgement segment is received that acknowledges the sent segment and estimates the end to end round-trip delay between the communicating hosts.
  • step 230 RTT is calculated that the Congestion and Competition Detection steps 250 are performed. These steps are only performed, if the TCP congestion control is in Slow Start or Congestion Avoidance (or an equivalent of these modes), but not in the Fast Recovery or Fast Retransmit modes. This is illustrated by a decision step 240.
  • RTT is not necessarily calculated for all new acknowledgements (ACKs). Nevertheless, also in such case, the congestion and competition detection may be performed in the same step 250, and may use the most recent RTT value obtained from a previous (e.g. most recent) new ACK, for which the RTT was calculated.
  • SND.UNA is set in Step 220, before the congestion and competition Detection 250. However, SND.UNA could be also set after step 250, in which case step 250 will use the acknowledgement sequence in the New ACK instead of SND.UNA.
  • Step 260 includes all remaining processing logic required for a new ACK by a TCP implementation. Such processing is not part of the present invention and is also not required thereby.
  • Step 270, Transmitting Data while respecting the Congestion Window size includes evaluating the Send Window (as a function of the Congestion Window, the Received Window and output buffer limits in the sending host), and then sending out TCP segments whose end sequence number is lower than the sum of SND.UNA and the Send Window. These segments would come from the queue of TCP segments waiting to be transmitted in the direction opposite to that of the received new ACK. A part of the new ACK processing after RTT estimation 260 could alternatively be performed after step 270.
  • the congestion control implementation may already include a method to determine the transition from Slow Start to Congestion Avoidance before losses occur, i.e. a method for setting the value of the Slow Start Threshold (ssthresh) to that of the current Congestion Window during Slow Start, as a function of measurements during that same Slow Start phase.
  • a method for setting the value of the Slow Start Threshold (ssthresh) to that of the current Congestion Window during Slow Start as a function of measurements during that same Slow Start phase.
  • the whole of that existing method which will normally be invoked upon receipt of a new ACK segment, may be replaced by the corresponding steps for congestion detection in Slow Start according to an embodiment of the invention.
  • the congestion detection is based on measuring a transmission rate and a flight size.
  • the transmission rate may be either the sending rate or the acknowledgement rate.
  • Sending Rate is the rate at which new TCP user data is being sent.
  • Acknowledgement Rate is the rate at which TCP user data is being acknowledged.
  • the acknowledgement rate is used as transmission rate
  • the sequence number and the SND.UNA value are preferably recorded at the time of sending the segment, so that when an acknowledgement for the sent data is received, the recorded SND.UNA value can be read, indicating the value of SND.UNA at the time when the acknowledged data was sent. This value can be recorded for all new data segments being sent or just for a subset of them, but in the latter case, precision in calculating the acknowledgement rate would be lost.
  • Figure 3 shows one possible embodiment of the methods of the invention, implementing any one of the five possible congestion detection steps (with their respective associated actions).
  • a method is provided for congestion control in a communication protocol employing acknowledged communication in which a transmitting node transmits data to a receiving node and the receiving node acknowledges receipt of the data, wherein a congestion window specifies a maximum amount of unacknowledged data which the transmitting node may transmit before reception of a positive acknowledgement for all or part of that data.
  • the method includes measuring 310 a flight size indicating an amount of data sent by the transmitting node and not yet acknowledged by the receiving node; measuring 320 a transmission rate, the transmission rate corresponding to a same time instant as the flight size; determining 350 a trend of the flight size; determining 360 a trend of the transmission rate; and judging 370 whether or not there is a congestion according to the determined trend of the transmission rate and the trend of the flight size. If in the judging step 370 (detection step), the congestion was positively detected, the step of modifying the congestion window 390 follows. The data to be transmitted are then transmitted in step 270, while respecting the congestion window size.
  • Fig. 3 shows a possible implementation of the congestion and competition detection step 250 to be executed as part of the processing of a new ACK, after the calculation 230 of RTT.
  • Fig. 3 there is one more step that is executed after the steps described in Fig. 3 (see Fig. 2 ), namely the transmission of the data while respecting the congestion window size 270, already described above.
  • the first two steps of the embodiment of Fig. 3 are: Measuring Flight Size 310 and Measuring Transmission Rate 320.
  • the step of Measuring Flight Size 310 obtains the flight size further used for detecting the congestion and possibly for distinguishing between the congestion with and without unfair competition.
  • the flight size may be measured (calculated) as follows. If using the sending rate for the data transmission rate, an appropriate sample of the Flight Size can be obtained by subtracting SND.UNA from SND.NXT at the time the new ACK is being processed. As a final result of processing a New ACK, normally new segments will be sent. Thus, instead of SND.NXT, it would be possible to use an estimate of what SND.NXT will be once those new segments have been sent out.
  • SND.NXT is a TCP connection state parameter, the first sequence number of the next unsent segment.
  • Flight Size is to be estimated at the time when the segment being acknowledged was sent.
  • the implementation of the invention may keep a record of (i.e. store) the SND.UNA parameter at the time the respective segments are sent, for all sent unacknowledged segments, or for a subset of them. Then, a sample of the number of bytes sent but not yet acknowledged can be obtained by subtracting SND.UNA stored at the time when the last segment being acknowledged was sent from SND.UNA at the time of processing the new ACK.
  • the SND.UNA (stored) at the time the last segment being acknowledged was sent
  • the SND.UNA at the time the first segment that is being acknowledged was sent is used for the calculation (which can be different from the last segment being acknowledged, especially when using delayed acknowledgements).
  • the SND.UNA parameter at the time segments are sent when several segments are sent consecutively as a response to a received ACK segment, as a burst, to also store how many consecutive segments are being sent, so that when calculating the flight size, the same flight size can be assigned to the acknowledgement of any of the segments sent out as a burst, namely the flight size that would correspond to the last segment sent as part of the burst.
  • the Congestion Window parameter cwnd itself can be used as an estimate of the flight size, which would produce results similar to using the acknowledgement rate.
  • the Congestion Window may be a rather inaccurate representation of the flight size.
  • the measured transmission rate may be calculated in step 320 as the above calculated flight size divided by the measured RTT.
  • Flight Size the data sending rate or the acknowledgement rate is obtained.
  • Using the data acknowledgement rate yields better results in congestion and unfair competition detection, because it is a better reflection of the end-to-end transmission rate and because it is a less noisy measurement, but it may also be more computationally and memory-intensive.
  • both options may be used for implementing the invention.
  • the sending rate and the acknowledgement rate are very similar. However, when there are fast variations in transmission rate, as in Slow Start, the sending rate may be quite different from the acknowledgement rate.
  • the evolution in time, i.e. the trend, of both the flight size and of the transmission rate may be determined (steps 350 and 360).
  • Fig. 7 illustrates a possible implementation of the determination of the trend in the flight size (step 350).
  • flightSizeTrend represents a variable that may be assigned the following values: "INDETERMINATE”, “UP”, “DOWN” and “STABLE”, and is initially set to a default value of "INDETERMINATE” in step 710.
  • the flightSizeMeasured variable stores the measured value from step 310, and if it is less than a configurable value flightSizeMin in step 715, the flight size trend estimation procedure is left (with flightSizeTrend set to "INDETERMINATE").
  • FlightSizeMin may be set to a value below which it may not be advantageous to determine the trend or to detect congestion, for example 20*MSS.
  • MSS stands for maximum segment size, corresponding to the largest amount of data (usually in octets/bytes), which can be transmitted or received in a single TCP segment (i.e. within the TCP payload).
  • variable congType represents the type of congestion that is being detected, and may have the values: "congSS” (for detection of congestion without unfair competition in Slow Start), “congNoUnfair” (for detection of congestion without unfair competition in Congestion Avoidance), “congUnfair” (for detection of congestion with unfair competition in Congestion Avoidance), “afterSS” (for detection of congestion with unfair competition after switching from Slow Start to Congestion Avoidance) and “afterCwnd” (for detection of congestion with unfair competition after reducing the congestion window due to a congestion detection in Congestion Avoidance).
  • congType for detection of congestion without unfair competition in Slow Start
  • CongNoUnfair for detection of congestion without unfair competition in Congestion Avoidance
  • CongUnfair for detection of congestion with unfair competition in Congestion Avoidance
  • afterSS for detection of congestion with unfair competition after switching from Slow Start to Congestion Avoidance
  • afterCwnd for detection of congestion with unfair competition after reducing the congestion window due to a congestion detection in Congestion Avoidance
  • the present invention can also be implemented for only one of the above types of congestion, in which case only the presence/absence of that particular congestion type would be detected.
  • congType is initially set to "congSS" when the connection starts in Slow Start, and then it is appropriately changed to other values, according to the current state of the network.
  • Fig. 7 which illustrates an embodiment of step 350, it is assumed that congType is already set. If it is set to either "afterCwnd” or "afterSS” in step 720, it will then compare the SND.UNA value with the seqRef variable in step 725, which stores the sequence number of the next unsent contiguous segment at the time a congestion was detected and either the congestion window was thereby reduced or the congestion control state was thereby switched from Slow Start to Congestion Avoidance. If SND.UNA is lower than seqRef, the flight size trend estimation procedure is left (with flightSizeTrend set to "INDETERMINATE").
  • step 735 checks the value of validRef, a variable that indicates whether or not a valid reference value has already been chosen as reference for the trend estimation. If validRef is "FALSE” and congType is verified in step 740 to be "afterCwnd", then, in step 750, the flightSizeRef variable (the flight size value to be used as reference for the trend estimation) will be set to the value stored in flightSizePrev, i.e., to the flightSizeMeasured value right before the congestion detection indicated by seqRef.
  • step 740 the flightSizeRef variable will be set to the current flightSizeMeasured value.
  • step 735 validRef is "TRUE" (i.e.
  • step 760 the flight size trend estimation procedure is also left (with flightSizeTrend set to "INDETERMINATE").
  • seqTest is a variable storing a sequence number after which the trend determination will be performed in case congType is either "afterCwnd" and "afterSS".
  • step 720 If in step 720 congType is not "afterCwnd” or "afterSS”, validRef is evaluated in step 795, and if varidRef is not TRUE, then the flightSizeRef variable will be set to the current flightSizeMeasured value in step 745 and the flight size trend estimation procedure is left (with flightSizeTrend set to "INDETERMINATE").
  • step 765 will be performed, to calculate the variable flightSizeThresh, which stores a threshold value based on which the trend in flight size will be determined, measured in bytes in this particular implementation.
  • This threshold value may be calculated as a fixed ratio of the reference flight size (stored in flightSizeRef), with 1/8 being an appropriate value, also because a division by 8 can be performed as a binary record shift.
  • the resulting value can then be clamped between a minimum value, flightsizeDiffMin (5*MSS, for example, would be appropriate), and a maximum value, flightsizeDiffMax (15*MSS, for example, would be appropriate),
  • Step 770 checks whether the current value of the flight size is lower than its reference value by more than the threshold calculated above, and if it is, in step 775, the variable flightSizeTrend is set to "DOWN" and the current values are taken as new reference values for both the flight size (stored in flightSizeRef) and for the transmission rate (stored in the rateRef variable).
  • the rateRef variable is analogous to the fligthSizeRef variable, but used for the transmission rate reference value, which is used in step 360 to calculate the trend of the transmission rate.
  • step 780 checks whether the current value of the flight size is higher than its reference value by more than that same threshold, and if it is, in step 780, the variable flightSizeTrend is set to "UP”. Finally, if step 780 determines that the current value of the flight size is not higher than its reference value by more than the aforementioned threshold, then the variable flightSizeTrend is set to "STABLE" in step 790.
  • the flight size trend estimation procedure is left (with flightSizeTrend set to its determined value, indicating either "UP” meaning increasing flight size trend or "DOWN” meaning decreasing flight size trend or "STABLE” meaning stable, i.e. substantially unchanged, flight size trend).
  • the trend determined in this way may be further used in step 370 for detecting the congestion.
  • Fig. 8 illustrates a possible implementation of the determination of the trend in the transmission rate (step 360).
  • the first step, step 330 Filter Transmission Rate is optional only if step 340 Filter Rate Gradient is performed (i.e. either step 330 or step 340, or both, must be implemented, to counter the variability in measured Transmission Rate).
  • Which filtering step is used will depend on which metric is used for the transmission rate gradient.
  • the Transmission Rate both the sending and the acknowledgement rate
  • a widely used and computationally efficient low-pass filter is a basic average estimator of the class called recursive prediction error algorithms, also called Exponentially Weighted Moving Average (EWMA).
  • EWMA Exponentially Weighted Moving Average
  • the sign "•" means multiplication, which is in Figures also denoted as "*". For example, value of 1/16 is an appropriate value for g when samples are obtained for every reception of a new ACK, but a higher value may be beneficial for less frequent samples. Nevertheless, the present invention is not limited to such values. Rather, a value may be selected which results in a most robust congestion control mechanism by testing. Other filtering algorithms could also be used, such as second degree low-pass filters.
  • RateFilterRestart represents a variable that may have either value "True” or "False”.
  • the variable rateMeasured is a variable that stores the transmission rate previously measured in step 320.
  • any kind of filtering may be applied, including for instance any (weighted) average filtering or any low-pass filtering.
  • rateTrend represents a variable that may be assigned the following values: "INDETERMINATE”, “UP”, “DOWN” and “STABLE”, and is initially set to a default value of "INDETERMINATE” in step 810.
  • the rateFiltered variable stores the filtered value of the transmission rate in step 330 (or the measured value from step 320 if no filtering is used).
  • step 820 compares the SND.UNA value with the seqRef variable. If SND.UNA is lower than seqRef, the transmission rate trend estimation procedure is left (with rateTrend set to "INDETERMINATE").
  • step 825 checks the value of validRef. If validRef is "FALSE" then, in step 830, the rateRef variable (the transmission rate value to be used as reference for the trend estimation) will be set to the value stored in ratePrev, i.e., to the rateFiltered value right before the congestion detection indicated by seqRef. It is assumed that ratePrev stores the previous filtered rate value, if using step 330 to filter rate measurements, or the previous rate measurement if not using step 330 to filter rate measurements (in which case, the ratePrev variable should be updated right before step 320 collects a new rateMeasured value).
  • the restartRateFilter and restartGradFilter variables will be set to TRUE, as a way to re-start the filtering once the reference transmission rate has been chosen for trend estimation; the validRef variable, will be set to "True” to indicate that both the transmission rate and flight size reference values are taken, and the SND.NXT parameter will be stored in the seqTest variable, since that is the sequence value at which the trend calculation will be performed when congType is either "afterCwnd" and "afterSS”.
  • the transmission rate trend estimation procedure is left (with rateTrend set to "INDETERMINATE”). If in step 825 validRef is "TRUE" (i.e. there is a valid reference transmission rate for trend determination), but SND.UNA is less than seqTest in step 835, the transmission rate trend estimation procedure is also left (with rateTrend set to "INDETERMINATE").
  • step 815 congType is not "afterCwnd” or "afterSS”, validRef is evaluated in step 840, and if varidRef is not TRUE, then the rateRef variable will be set to the current rateFiltered value in step 845, while the validRef variable will be set to "True” to indicate that both the transmission rate and flight size reference values are taken, and the transmission rate trend estimation procedure is left (with flightSizeTrend set to "INDETERMINATE").
  • step 850 will calculate the transmission rate gradient and store it in the rateGradient variable.
  • step 340 Filter Rate Gradient
  • step 340 Filter Rate Gradient
  • step 855 will calculate the variables gradientThreshUp and gradientThreshDown, which store threshold values based on which the trend in transmission rate will be determined.
  • the transmission rate gradient in step 850 may be calculated in different ways and this will influence how it is filtered in step 340 and the calculation of threshold for trend determination in step 855.
  • rateGradient rateFiltered ⁇ rate Ref
  • the threshold value for upward trend determination, gradientThreshUp may be calculated as a fraction (e.g. 1 ⁇ 2) of the ratio of the threshold flight size to the reference flight size (i.e. flightSizeThresh/flightSizeRef), applied to the reference transmission rate (stored in rateRef). The resulting value can then be forced to be smaller than a certain ratio of the reference transmission rate (e.g.
  • the threshold to determine if there is a downward trend, gradientThreshDown, may advantageously be calculated as twice gradientThreshUp, but with negative sign. Twice, or some other factor greater than 1, in order to have a lower probability of erroneously detecting a downward trend.
  • the gradient thresholds can be linked to the variability in the measured transmission rate.
  • an estimate of that variability is required, such as the mean absolute deviation.
  • a multiple e.g. 2 for upward trend threshold and 4 for downward trend threshold
  • Yet another alternative to calculate the gradient could be to determine the change in transmission rate associated with every change in flight size. Since there may be several contiguous new ACK's with the same measured flight size and different measured transmission rates (in congestion avoidance, for example, the actual congestion window is actually increased once every congestion window bytes, which should normally be reflected in the same way in the flight size), an initial filtering would include determining the mean transmission rate value for every flight size value (these mean values could be stored in a rateFiltered variable, for example).
  • the ratePrev value would correspond to the mean transmission rate calculated when the flight size had the flightSizePrev value
  • rateFiltered would correspond to the mean transmission rate calculated when the flight size is flightSizeMeasured.
  • This value would be calculated once for every change in flight size.
  • this gradient could be normalized by dividing it by (ratePrev/ flightSizePrev). So normalized, a value of 1 would correspond to a situation in which the transmission rate increases, as a percent ratio, the same as the flight size, while a value of 0 would indicate a stable trend and a negative value a downward trend.
  • a gradient calculated in this way will show a relatively large variability, so a Filter Rate Gradient step 340 would be very advantageous in reducing the observed variability.
  • This step 340 could be implemented as shown in Fig. 6 , which directly replicates the transmission rate filtering embodiment shown in Fig. 5 .
  • appropriate threshold values would be 0.5 for gradientThreshUp and -1 for gradientThreshDown.
  • Step 860 checks whether the current value of the (possibly filtered) gradient of the transmission rate is lower than its gradientThreshDown threshold calculated above, and if it is lower, in step 865, the variable rateTrend is set to "DOWN". If step 860 determines that the filtered transmission rate is not lower than its gradientThreshDown threshold calculated above, then step 870 checks whether the current value of the filtered transmission rate is higher than its gradientThreshUp threshold, and if it is, in step 875, the variable rateTrend is set to "UP”. Finally, if step 870 determines that the current value of the filtered transmission rate is not higher than its gradientThreshUp threshold, then the variable rateTrend is set to "STABLE" in step 880. After any of steps 865, 875 and 880, the transmission rate trend estimation procedure is left (with rateTrend set to its determined value, UP, DOWN or STABLE).
  • the trend of flight size or transmission rate may be estimated by comparing gradient of the flight size or transmission rate with a threshold.
  • the gradient may be calculated, for instance, as a difference between the current and the previous value of the flight size or the transmission rate, respectively.
  • the difference may be calculated between the filtered previous value and the filtered current value, which results in variation-robust results.
  • the differences may also be calculated between the previous filtered value and the measured current value, which may lead to faster reactions to temporal changes.
  • the "previous value” is beneficially an immediately preceding value (filtered or measured). However, this is not meant to limit the present invention and the previous value may also be any preceding value. Moreover, the calculation of the trend may also be performed by taking into account more than one preceding measured or filtered values, which may provide the advantage of a more robust trend estimation.
  • Fig. 9 illustrates a possible implementation of the congestion detection (corresponding to step 370).
  • congType is "afterSS” (in step 905), it will add MSS bytes to the shadowCwnd variable.
  • the shadowCwnd variable stores the value that the congestion window would have reached if it had remained in Slow Start (congType set to "afterSS” means it has recently left the Slow Start state), to be able to revert to it if necessary.
  • congType is set to "afterCwnd” in step 910 or if it is "afterSS” after step 995, (i.e.
  • step 915 will check if the previously determined (in step 350) trend of flight size (flightSizeTrend) is "STABLE” or “UP”, and if it is not (i.e. if it is "DOWN” or "UNDETERMINATE"), then step 930 will set the congestion Detected variable to False and end the congestion detection procedure with the result of negative congestion detection (no congestion detected) . If flightSizeTrend is "STABLE” or "UP” in step 915, step 920 will then set the validRef variable to False, so that subsequent trend determination steps (steps 350 and 360) are forced to set new reference values.
  • step 935 will set the congestion Detected variable to "True” and end the congestion detection procedure with the result of positive congestion detection (congestion in the network detected). Otherwise, if rateTrend is not "DOWN” in step 925 (i.e. if it is UP, STABLE or UNDETERMINATE), then step 930 will set the congestionDetected variable to False and end the congestion detection procedure (negative detection).
  • step 945 will check if the previously determined (in step 350) trend of flight size (flightSizeTrend) is "UP”, and if it is not (i.e. if it is "STABLE”, “DOWN” or "UNDETERMINATE") then step 960 will set the congestion Detected variable to False and end the congestion detection procedure.
  • step 950 will then set the validRef variable to False, so that subsequent trend determination steps (steps 350 and 360) are forced to set new reference values.
  • step 965 will set the congestion Detected variable to "True” and end the congestion detection procedure. Otherwise, if rateTrend is not "DOWN” in step 955 (i.e. if it is UP, STABLE or UNDETERMINATE), then step 960 will set the congestion Detected variable to False and end the congestion detection procedure (negative detection).
  • step 970 will check if the previously determined (in step 350) trend of flight size (flightSizeTrend) is "UP”, and if it is not (i.e. if it is "STABLE”, “DOWN” or "UNDETERMINATE") then step 990 will set the congestion Detected variable to False and end the congestion detection procedure. If flightSizeTrend is "UP” in step 970, step 975 will then set the validRef variable to False, so that subsequent trend determination steps (steps 350 and 360) are forced to set new reference values.
  • flightSizeTrend is "UP" in step 970
  • step 975 will then set the validRef variable to False, so that subsequent trend determination steps (steps 350 and 360) are forced to set new reference values.
  • step 975 if the previously determined (in step 360) trend of the transmission rate (rateTrend) is "STABLE” in step 980, then step 985 will set the congestion Detected variable to "True” and end the congestion detection procedure. Otherwise, if rateTrend is not "STABLE” in step 980 (i.e. if it is UP, DOWN or UNDETERMINATE), then step 990 will set the congestion Detected variable to False and end the congestion detection procedure.
  • Figure 9 illustrates the detection of a congestion situation and takes into account all above mentioned kinds of congestion (congestions in the respective Slow Start or Congestion Avoidance, congestion with or without competition).
  • the present invention may also implement only detection and handling of one or a subset of these congestion types. Accordingly, only subsets of steps shown in Fig. 9 may be implemented. For instance, for detecting congestion without unfair competition, only steps 970-990 may be implemented. For instance, for detecting congestion with unfair competition, only steps 945-965 may be implemented.
  • the step 370 of detecting whether or not there is congestion includes detecting whether or not there is a congestion with unfair competition.
  • the congestion with unfair competition is then positively detected (step 965) when the trend of the flight size is increasing (step 945) and the trend of the transmission rate is decreasing (step 955), and negatively detected otherwise.
  • the step of detecting whether or not there is a congestion 370 may include detecting whether or not there is a congestion without unfair competition; and the congestion without unfair competition may be positively detected when the trend of the flight size is increasing (step 970) and the trend of the transmission rate is substantially equal (step 980), and negatively detected otherwise.
  • Fig. 10 illustrates a possible implementation of the congestion window modification (corresponding to step 390).
  • this step is executed only if some kind of congestion has been previously detected in the congestion detection step 370.
  • Step 390 then contains the actions to be performed in each of the five possible congestion types detected by step 370 and specified in the congType variable. If congType is "congSS" in step 1010 (i.e. if a congestion without unfair competition has been detected in slow start, for instance in step 935), then step 1015 will set the slow start threshold variable (ssthresh) to the current value of the congestion window (cwnd) thereby switching the congestion control state to Congestion Avoidance.
  • congType is "congSS" in step 1010 (i.e. if a congestion without unfair competition has been detected in slow start, for instance in step 935)
  • step 1015 will set the slow start threshold variable (ssthresh) to the current value of the congestion window (cwnd) thereby switching the congestion control state to Congestion Avoidance
  • step 1020 will set the shadowCnd variable to cwnd, too, so that this variable can be used to track what the congestion window could have been if the congestion had stayed in slow start (in the described embodiments, the shadowCwnd is subsequently updated in step 370).
  • step 1015 is only necessary if after the detection of congestion without unfair competition in slow start, congType is set to "afterSS" to detect if there is congestion with unfair competition right after the switch from Slow Start to Congestion Avoidance.
  • step 1015 would not be required in the general case of embodiment illustrated in Fig.3 but would be required for the more particular embodiment of Fig. 4 (described below in detail).
  • step 1030 will reduce the congestion window by an amount of deltaCwnd. That amount in bytes (deltaCwnd) may be a fixed proportion of the congestion window (for example, 20%) or may be a proportion of the flight size threshold used in determining the trend of the flight size (flightSizeThresh). In the latter case, it is advisable for it to be somewhat larger (e.g. 1.2 x flightSizeThresh), so if a congestion has been detected, the congestion window will be reduced beyond the real congestion trigger, thus allowing it to oscillate around that congestion trigger point in successive rounds of congestion window growth and reductions due to congestion detection.
  • deltaCwnd may be a fixed proportion of the congestion window (for example, 20%) or may be a proportion of the flight size threshold used in determining the trend of the flight size (flightSizeThresh). In the latter case, it is advisable for it to be somewhat larger (e.g. 1.2 x flightSizeThresh), so if a congestion has been detected, the congestion
  • Step 1035 follows either step 1020 or step 1030, but it is only necessary if after the detection of congestion without unfair competition, congType is set to either "afterSS” or “afterCwnd", to detect if there is congestion with unfair competition right after the switch from slow start to congestion avoidance or after the congestion window reduction, respectively. Thus, step 1035 would not be required in the embodiment illustrated in Fig. 3 but would be implemented for that of Fig. 4 .
  • Step 1035 includes setting the variable seqRef to SND.NXT, which is the sequence number at which the reference values for flight size and transmission rate will be chosen in determining their trends (actually, the values just prior to seqRef are taken, except in the case of the flight size when congType is afterCwnd, when the value right after seqRef is taken, as can be seen in Fig. 7 and in Fig. 8 ).
  • step 1045 will set the slow start threshold variable (ssthresh) to a larger value, for example 3 times the current value of the congestion window (cwnd), thereby switching the congestion control state to slow start.
  • slow start threshold variable for example 3 times the current value of the congestion window (cwnd)
  • step 1055 will set the congestion window to the value of the shadowCwnd variable (which reflects the value that would have been attained by cwnd if the connection had stayed in slow start).
  • Step 1055 will also set the slow start threshold variable (ssthresh) to a larger value, for example 3 times the value of the congestion window (cwnd), thereby switching the congestion control state to slow start.
  • step 1065 will add to the congestion window the value previously stored in the deltaCwnd variable reflecting the reduction in cwnd when congestion was detected at a prior time. In this way, the reduction in cwnd is "undone".
  • restartGradFilter, restartRateFilter and validRef variables should be initialized to "FALSE" whenever the connection starts or there is an RTO time-out or an inactivity time-out.
  • congType should be set at the start of the connection to whatever congestion detection type is required.
  • Figure 10 shows handling of all above described congestion conditions, and in particular, to modification 390 of the congestion window.
  • the step of modifying the congestion window 390 includes increasing (step 1045) the slow start threshold upon positively detecting the congestion with unfair competition (step 1040), which will switch the connection to Slow Start.
  • step 10 exemplified the slow start threshold increase as multiplying the current congestion window by three.
  • this is only an example and the increase may be performed by a different multiplier or in another way.
  • the increase of the congestion window size upon positively detecting the congestion with unfair competition is beneficially performed if the transmitting node is in TCP congestion avoidance congestion control state and not performed otherwise.
  • the step 390 of modifying the congestion window may include reducing (step 1030) the congestion window size upon positively detecting (step 1025) the congestion without unfair competition.
  • the reduction of the congestion window size upon positively detecting the congestion without unfair competition, when using the TCP is performed if the transmitting node is in TCP congestion avoidance congestion control state ("No" in step 1010 for not being in Slow Start) and not performed otherwise.
  • the step 390 of modifying the congestion window may include changing from a current congestion control state to a new congestion control state upon positively detecting the congestion with unfair competition, wherein the current and the new congestion control states specify respectively a current and a new rule for increasing the congestion window size upon receipt of data acknowledgements when neither congestion nor data losses have been detected, and the new rule specifies a faster rate of increasing the congestion window size than the current rule.
  • the change from a current congestion control state to a new congestion control state upon positively detecting the congestion with unfair competition may be performed if the current congestion control state is TCP congestion avoidance and not performed otherwise, and the new congestion control state may correspond to the TCP slow start.
  • the change from a current congestion control state to a new congestion control state upon positively detecting the congestion with unfair competition is beneficially performed if the current congestion control state is TCP congestion avoidance and not performed otherwise, and the new congestion control state is TCP slow start.
  • the step 390 of modifying the congestion window may include changing from a current congestion control state to a new congestion control state upon positively detecting the congestion without unfair competition, wherein the current and the new congestion control states specify respectively a current and a new rule for increasing the congestion window size upon receipt of data acknowledgements when neither congestion nor data losses have been detected, and the new rule specifies a slower rate of increasing the congestion window size than the current rule.
  • TCP it is advantageous when the change from a current congestion control state to a new congestion control state upon positively detecting the congestion without unfair competition is performed if the current congestion control state is TCP slow start and not performed otherwise, and the new congestion control state is TCP congestion avoidance state.
  • Fig. 4 shows one possible embodiment of the methods of the invention, including all five types of congestion detection steps (with their respective associated actions) in a coordinated way.
  • Fig. 4 shows a possible implementation of the congestion and competition detection steps 250 which are assumed to be executed as part of the processing of a new ACK, after the calculation (230) of RTT.
  • the embodiment of the invention described in Fig.4 uses all the steps used in the embodiment described in Fig. 3 (310, 320, 350, 360, 370, 380 and 390) and adds more steps to manage the congType variable, which specifies the kind of congestion detection to use.
  • this embodiment will detect congestion without unfair competition when in Slow Start, and when such congestion is detected, the connection will be switched to congestion avoidance.
  • Right upon switching to Congestion Avoidance it will determine if there is then congestion with unfair competition with reference to the time when the switching occurred and if so, switch back to connection to Slow Start. Once in Congestion Avoidance, it will check for congestion without unfair competition, which will be treated by reducing the congestion window.
  • step 405 if the connection is in Slow Start, the congType variable is set to "congSS" in Step 410 to determine the detection of congestion without unfair competition in Slow Start. If not, the previous value of congType will remain. In both cases, the following two steps will be performed to measure the flight size, 310 and to measure the transmission rate, both of which can be implemented as described above (when describing those same steps for the embodiment illustrated by Fig. 3 ).
  • next steps will be those of determining the trend of flight size 350 (with one possible embodiment illustrated in Fig. 7 ), determining the trend of transmission rate 360 (with one possible embodiment illustrated in Fig. 8 ) and detecting congestion 370 (with one possible embodiment illustrated in Fig. 9 ), all of them already described in detail as part of the description of the embodiment in Fig. 3 .
  • step 380 the congestion Detected variable is True (i.e., that some kind of congestion was detected in step 370)
  • step 390 (with one possible embodiment illustrated in Fig. 10 ) will be performed to modify the congestion window appropriately.
  • step 455 will set congType to "congNoUnfair” (congestion detection without unfair competition in congestion avoidance). If, however, congType was "congSS” (congestion detection without unfair competition in slow start), step 460 will lead to step 465, in which congType will be set to "afterSS” (congestion detection with unfair competition after a transition to slow start).
  • step 470 will lead to step 475, in which congType will be set to "afterCwnd” (congestion detection with unfair competition after a reduction in congestion window).
  • step 415 will verify if SND.UNA is larger than seqTest in cases where congType is either "afterSS” or "afterCwnd", and if so set congType to "congNoUnfair” in step 420. This ensures that the detection of congestion with unfair competition with respect to specific events (the transition to Congestion avoidance and the reduction of congestion window due to congestion) is only performed up to the sequence number specified by seqTest, and after that the congestion detection mode will be switched to congestion detection without unfair competition in congestion avoidance.
  • step 415 determines that congType is "congNoUnfair” (i.e. congestion without unfair competition in congestion avoidance has been tested but not found)
  • step 430 will set the congType to "congUnfair” (congestion with unfair competition in congestion avoidance) and then run another congestion detection step 370 (with one possible embodiment illustrated in Fig. 9 ), with the same trend information but in this case checking for a different type of congestion. If step 435 verifies that this new congestion detection has been negative, then step 440 sets the congType back to "congNoUnfair", so that for the next new ACK congestion without unfair competitions will again be tested and if not found, congestion with unfair competition is tested.
  • step 435 finds that congestion with unfair competition was detected, then step 390 (with one possible embodiment illustrated in Fig. 10 ) will modify the congestion window appropriately an finally step 445 will set the congType back to "congNoUnfair", If, as in the embodiment of step 390 illustrated in Fig. 10 , the connection is switched to Slow Start in some cases, congType will not be changed to "congSS" until the next New ACK, when the step 405 detects the congestion is in Slow Start.
  • congType should be initialized to "congSS”
  • restartGradFilter, restartRateFilter and validRef variables should be initialized to "FALSE”, whenever the connection starts or there is an RTO time-out or an inactivity time-out.
  • step 360 could still be based on Fig. 8 , with a number of differences with respect to the implementation described above. The main difference would be that instead of the rateFiltered variable, a new variable called rateSample would have to be used, with rateSample being a vector or array of N (e.g. 5) samples of rateFiltered.
  • N e.g. 5
  • rateSample would have the last N values of rateFiltered, or a subset of the last values.
  • ratePrev the ratePrev variables
  • ratePrev the ratePrev variables
  • ratePrev the value of rateFiltered
  • ratePrev the value of ratePrev
  • Step 860 would check if the gradient in transmission rate is positive by checking if rateSample vector is greater than the vector resulting from the sum of gradientThreshUp to each of the N components of rateRef.
  • Step 870 would check if the gradient in transmission rate is negative by checking if rateSample vector is smaller than the vector resulting from subtracting gradientThreshUp from each of the N componenents of rateRef.
  • the test is easily performed by assigning a rank to each of the N values of the two vectors being compared, that is, ordering the 2*N values in sequence and assigning them their order number (with ties assigned the same number). Then, taking the vector/array with the lower median value, for each of its N components, count the number of observations in the other vector/array that have a smaller rank (count a half for any that are equal to it). The sum of these counts is the U statistic.
  • a U value of 5 or less means that the sample (vector/array) in this case is lower than the other, while a value of 20 or more means that the sample (vector/array) in this case is higher than the other.
  • An alternative way to determine the trend of the transmission rate is to use a statistical test to detect a trend in the transmission rate measurements and then use a metric produced by the test as the gradient, with the metric being then filtered (by accumulating or averaging it over time, for example).
  • One possible such test, to detect an upward trend would be to fit two models to the measured data, one with the expected behaviour without congestion and the other with the expected behaviour without congestion. Then, a metric that compares how well the two models fit the measured data can be used as a kind of gradient from which the trend in transmission rate is derived.
  • every transmission rate measurement may be compared against a model that predicts it to be the same as the previous value (that is, a random walk model, which could be expected from a congestion situation), and then compared against a model that predicts an increment in transmission rate from the last value in the same proportion as the measured flight size increment (which could be expected in a non-congestion situation).
  • a model that predicts it to be the same as the previous value that is, a random walk model, which could be expected from a congestion situation
  • a model that predicts an increment in transmission rate from the last value in the same proportion as the measured flight size increment which could be expected in a non-congestion situation.
  • Fig. 11 A possible implementation of this method is illustrated in Fig. 11 , which is identical to Fig. 8 except in a few aspects that will now be described.
  • the gradient described in the previous paragraph can only be used to detect an upward trend in transmission rate, so another gradient will be required for testing if the trend is downward.
  • the gradient produced by step 851 will be a two-valued variable: it will produce a gradient for the downward trend test (to be stored in a variable called rateGradientDown) and a gradient for the upward trend test (to be be stored in a variable called rateGradientDown).
  • step 856 will produce one threshold value for the upward trend test (to be stored in the gradientThreshUp variable) and another for the downward trend test (to be stored in the gradientThreshDown variable).
  • rateGradientUp rateMeasured ⁇ ratePrev ⁇ 2 / rateMeasured ⁇ ratePrev • 1 + flightSizemeasured ⁇ flightSizePrev / flightSizePrev ⁇ 2
  • rateGradientDown rateFiltered ⁇ rate Ref
  • step 895 Filter rateGradienUp and rateGradienDown, could use an exponentially weighted moving average of both gradients (as described above).
  • Other filtering methods would be possible for rateGradienUp, like accumulating independently the sum of squared residuals for each of the two models, and then producing an updated filtered gradient by dividing the accumulated sums of squared residuals.
  • the threshold used for the downward trend calculation is as already used in previously described exemplary embodiments of step 360, while the threshold of 2 used for gradientThreshUp would be reached when the squared residuals (the errors) in the model predicting congestion are larger (twice as large) than the squared residuals (errors) in the model without congestion.
  • steps 861 and 871 will compare the threshold and make the trend determination in almost the same way as previously described for other embodiments of step 360, except that there would be a specific gradient used for the upward trend determination and another for the downward trend determination.
  • An embodiment of the present invention includes an apparatus which is configured to execute any of the above described methods.
  • An example of such apparatus is illustrated in Figure 12 .
  • an apparatus for controlling, in a transmitting node (such as sending host 110, 111 or proxy 130), congestion in a data communication protocol employing acknowledged communication in which the transmitting node transmits data to a receiving node and the receiving node acknowledges receipt of the data, wherein a congestion window specifies a maximum amount of unacknowledged data which the transmitting node may transmit before reception of a positive acknowledgement for all or part of that data.
  • a transmitting node such as sending host 110, 111 or proxy 130
  • a congestion window specifies a maximum amount of unacknowledged data which the transmitting node may transmit before reception of a positive acknowledgement for all or part of that data.
  • the apparatus comprises a flight size measuring unit 1210 for measuring a flight size indicating an amount of data sent by the transmitting node and not yet acknowledged by the receiving node; a transmission rate measuring unit 1220 for measuring a transmission rate, the transmission rate corresponding to a same time instant as the flight size; a flight size trend calculation unit 1250 for determining a trend of the flight size; a transmission rate trend calculation unit 1260 for determining a trend of the transmission rate; a congestion detection unit 1270 for detecting whether or not there is a congestion according to the determined trend of the transmission rate and the trend of the flight size; a congestion handling unit 1280 for, upon positive detection of the congestion by the congestion detection unit, modifying the congestion window; and a transmission unit 1290 for transmitting data while respecting the congestion window size.
  • a flight size measuring unit 1210 for measuring a flight size indicating an amount of data sent by the transmitting node and not yet acknowledged by the receiving node
  • a transmission rate measuring unit 1220 for measuring a transmission rate, the transmission rate corresponding to
  • such apparatus advantageously includes a TCP Logic Processing Unit 1200 and a Congestion and Competition Detection Unit 1295 which at least logically (and possibly also physically, i.e. by implementing them by a single processing device) encompasses the above mentioned units.
  • the TCP Logic Processing Unit 1200 includes the logic required to send and receive data using the TCP protocol.
  • the Reception Unit 1205 in the TCP Logic Processing Unit comprises the TCP functionality for receiving TCP segments while the Transmission Unit 1290 comprises the TCP functionality for sending TCP segments, in both cases according to the TCP standards, like the IETF RFC 793 and others cited above.
  • the Transmission Unit in particular, also includes the enforcement of the TCP Congestion Window and the TCP Send Window, so that packets are transmitted only when those parameters permit.
  • the Congestion and Competition Detection Unit 1295 comprises the other elements in the invention, apart from the above mentioned Transmission Unit: a Flight Size Measuring Unit 1210, a Transmission Rate Measuring Unit 1220, a Flight Size Trend Calculating Unit 1250, a Transmission Rate Trend Calculating Unit 1260, a Congestion Detection Unit 1270 and a Congestion Handling Unit 1280.
  • the TCP Logic Processing Unit 1200 invokes the functionality of the Congestion and Congestion and Competition Detection Unit for every New ACK (as previously defined) received by the Reception Unit for connections in the Slow Start or Congestion Avoidance states, possibly including the value of the SND.UNA variable at the time the data segment being acknowledged by the New ACK.
  • the Congestion and Congestion and Competition Detection Unit has access (via memory, for example) to the current values of the SND.NXT, SND.UNA, last measured RTT, cwnd and ssTrhesh variables of the TCP connections in the TCP Logic Processing Unit, including the possibility of changing the stored values of the cwnd and ssThresh. Additionally, the internal units of the Congestion and Competition Detection Unit may communicate among themselves through variables stored in memory.
  • the TCP Logic Processing Unit 1200 and the Congestion and Congestion and Competition Detection Unit 1295 may be implemented as part of the kernel of an operating system 1310, as illustrated in Fig. 13 .
  • the operating system kernel includes the network interface drivers and software libraries that enable sending and receiving data packets to and from other systems via data networks.
  • the TCP Processing Logic Unit 1200 could be the TCP functionality in the kernel of most operating systems (e.g. Linux, Windows, BSD, OSX, Android, IOS), adapted to invoke the Congestion and Competition Detection Unit 1295 for New ACKs as described above, allowing access to the SND.UNA, SND.NXT, last measured RTT, cwnd and ssThresh also as described above.
  • the Transmission Unit 1290 in the Processing Logic Unit 1200 would require no adaptations in that case, except for storing the value of the SND.UNA variable when segments are sent out.
  • the TCP Logic Processing Unit 1200 and the Congestion and Congestion and Competition Detection Unit 1295 may be implemented not being part of the kernel of an operating system 1410, as illustrated in Fig. 14 , in "user-space".
  • the apparatus would have an operating system kernel, including the network interface drivers and software libraries that enable sending and receiving data packets to and from other systems via data networks.
  • the TCP Processing Logic Unit 1200 in this case implements the TCP protocol in "user-space" (outside the kernel), including a Transmission Unit 1290 to send out data segments according to the TCP protocol, i.e. respecting a Congestion Window and a Send Window for the connection, as previously described, via System Libraries 1420 that interface the communication with the operating system kernel.
  • the TCP Processing Logic Unit 1200 in this case stores the value of the SND.UNA variable when segments are sent out.
  • the implementation of the units comprised in the Congestion and Congestion and Competition Detection Unit 1295 may be realized by configuring the execution of the steps described above, as part of the exemplary embodiments of the methods of this invention.
  • the Flight Size Measuring Unit 1210 is configured to execute step 310, Measuring the Flight Size
  • the Transmission Rate Measuring Unit 1220 is configured to execute step 320, Measuring the Transmission Rate
  • the Flight Size Trend Calculation Unit 1250 is configured to execute step 350, Determining the Trend of Flight Size
  • the Transmission Rate Trend Calculation Unit 1260 is configured to execute step 360, Determining the Trend of Transmission Rate
  • the Congestion Detection Unit 1270 is configured to execute step 370, Detecting Congestion
  • the Congestion Handling Unit 1280 is configured to execute step 390, Modifying Congestion Window, but only in cases where the Congestion Detection Unit has detected congestion.
  • the present invention may be embodied in an apparatus as shown in Fig. 15 corresponding to a processing device comprising one or more processors 1510, a memory 1520 or a set of memories and one or more network interfaces 1530, all interconnected, with the memory storing instructions which, when executed by the processor, cause the processor to perform operations comprising the steps of any of the methods described above.
  • the processor(s) may be configured for receiving New ACK messages (step 220, described as part of an embodiment of the methods of the invention and illustrated in Fig. 2 ); processing of New ACK message, possibly including SND.UNA parameter updating (step 220, described as part of an embodiment of the methods of the invention and illustrated in Fig.
  • step 230 described as part of an embodiment of the methods of the invention and illustrated in Fig. 2
  • step 230 described as part of an embodiment of the methods of the invention and illustrated in Fig. 2
  • step 250 described as part of an embodiment of the methods of the invention and illustrated in Fig. 2
  • step 260 additional processing of New ACK message, as required by TCP updating
  • step 270 described as part of an embodiment of the methods of the invention and illustrated in Fig. 2
  • the 15 may store the instructions for performing any of the method described above in the memory 1520 or a plurality of such memories.
  • the instructions may then be executed by the processor 1510 or a plurality of such processors and the congestion control may be then performed on the basis of the communication such as acknowledgement and data transmitted or received via the network interface 1530 and evaluated (measured, filtered, processed to gain trends, test conditions) as described above.
  • the Congestion and Competition Detection may, for example, be implemented as previously specified in the detailed description of the embodiment described in Fig.3 . It may also, for example, be implemented as previously specified in the detailed description of the embodiment described in Fig.4 .
  • the processors used for the apparatus implementation may be, for example, general-purpose processors, microcontrollers, processors optimized for mobile phones or tablet devices, customized ASIC's, FPGA's or other similar devices.
  • the memory may be RAM, ROM, EPROM, EEPROM, hard disk, solid-state disk, or similar storage devices, or any combinations thereof.
  • the network interfaces may allow sending and receiving data to networks via wire-based ports (e.g. Ethernet, cable, fiber, ISDN, ADSL) or wireless ports (e.g. Wifi, WiMax, GPRS, UMTS, HSPA, LTE, microwave links).
  • the networks for through which the apparatus embodying the present invention may communicate are, for example, LAN, WAN, satellite, 3G wireless, 4G wireless, cable, backhaul, etc..
  • the apparatuses exemplified above implement TCP communications. However, but the present invention is not limited thereto and may also be used with any other communication protocols using acknowledged communication.
  • apparatuses as described above may be applied to endpoints such as hosts (illustrated in Figure 1A ) and also to proxy nodes (illustrated in Figure 1B ).
  • endpoints such as hosts (illustrated in Figure 1A ) and also to proxy nodes (illustrated in Figure 1B ).
  • proxy nodes illustrated in Figure 1B .
  • an apparatus embodying the invention would apply the congestion and completion detection and control to the sending functionality.
  • the present invention may be embodied as a data carrier storing a program including instructions capable of causing, when run on a computer, the computer to execute the steps of any of the above described methods.
  • the data carrier may be a portable data carrier such as flash memory connectable to a computer via USB interface, an optical medium such as a CD, DVD, BlueRay or other.
  • the data carrier may also be a hard drive or any other storage medium.
  • the present invention relates to congestion control for acknowledged communication over networks.
  • the congestion is detected on the basis of the trends of both the flight size and the transmission rate in order to adapt the congestion window in accordance with the detection result.
  • Such congestion detection enables, for instance, distinguishing between the congestion with or without unfair competition.
  • the measured transmission rate may be filtered to compensate for time variations.
  • the invention may be embodied within an end node or within a proxy.

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Databases & Information Systems (AREA)
  • Data Exchanges In Wide-Area Networks (AREA)
  • Communication Control (AREA)
EP20184668.0A 2014-04-23 2015-03-31 Appareil de régulation de l'encombrement de réseau sur la base des gradients de vitesse de transmission Active EP3742687B1 (fr)

Applications Claiming Priority (4)

Application Number Priority Date Filing Date Title
EP14382146 2014-04-23
PCT/EP2015/056962 WO2015161990A1 (fr) 2014-04-23 2015-03-31 Procédé et appareil de contrôle d'encombrement de réseau basé sur des gradients de débit de transmission
EP15713734.0A EP3135009B1 (fr) 2014-04-23 2015-03-31 Procédé et appareil de contrôle d'encombrement de réseau basé sur des gradients de débit de transmission
EP17206263.0A EP3319281B1 (fr) 2014-04-23 2015-03-31 Procédé et appareil de régulation de l'encombrement de réseau sur la base des gradients de vitesse de transmission

Related Parent Applications (3)

Application Number Title Priority Date Filing Date
EP17206263.0A Division EP3319281B1 (fr) 2014-04-23 2015-03-31 Procédé et appareil de régulation de l'encombrement de réseau sur la base des gradients de vitesse de transmission
EP17206263.0A Division-Into EP3319281B1 (fr) 2014-04-23 2015-03-31 Procédé et appareil de régulation de l'encombrement de réseau sur la base des gradients de vitesse de transmission
EP15713734.0A Division EP3135009B1 (fr) 2014-04-23 2015-03-31 Procédé et appareil de contrôle d'encombrement de réseau basé sur des gradients de débit de transmission

Publications (2)

Publication Number Publication Date
EP3742687A1 true EP3742687A1 (fr) 2020-11-25
EP3742687B1 EP3742687B1 (fr) 2024-09-11

Family

ID=50685847

Family Applications (3)

Application Number Title Priority Date Filing Date
EP17206263.0A Active EP3319281B1 (fr) 2014-04-23 2015-03-31 Procédé et appareil de régulation de l'encombrement de réseau sur la base des gradients de vitesse de transmission
EP15713734.0A Active EP3135009B1 (fr) 2014-04-23 2015-03-31 Procédé et appareil de contrôle d'encombrement de réseau basé sur des gradients de débit de transmission
EP20184668.0A Active EP3742687B1 (fr) 2014-04-23 2015-03-31 Appareil de régulation de l'encombrement de réseau sur la base des gradients de vitesse de transmission

Family Applications Before (2)

Application Number Title Priority Date Filing Date
EP17206263.0A Active EP3319281B1 (fr) 2014-04-23 2015-03-31 Procédé et appareil de régulation de l'encombrement de réseau sur la base des gradients de vitesse de transmission
EP15713734.0A Active EP3135009B1 (fr) 2014-04-23 2015-03-31 Procédé et appareil de contrôle d'encombrement de réseau basé sur des gradients de débit de transmission

Country Status (8)

Country Link
US (5) US10263894B2 (fr)
EP (3) EP3319281B1 (fr)
CN (1) CN106464598B (fr)
DK (1) DK3135009T3 (fr)
ES (2) ES2666426T3 (fr)
IL (1) IL248275B (fr)
NO (1) NO3135009T3 (fr)
WO (1) WO2015161990A1 (fr)

Families Citing this family (42)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9992088B1 (en) 2014-11-07 2018-06-05 Speedy Packets, Inc. Packet coding based network communication
US10320526B1 (en) 2014-11-07 2019-06-11 Strong Force Iot Portfolio 2016, Llc Packet coding based network communication
US9992126B1 (en) * 2014-11-07 2018-06-05 Speedy Packets, Inc. Packet coding based network communication
US10999012B2 (en) 2014-11-07 2021-05-04 Strong Force Iot Portfolio 2016, Llc Packet coding based network communication
US9825733B1 (en) 2014-11-07 2017-11-21 Speedy Packets, Inc. Packet coding based network communication
US9979664B2 (en) 2015-07-07 2018-05-22 Speedy Packets, Inc. Multiple protocol network communication
US10715495B2 (en) * 2015-12-15 2020-07-14 Nicira, Inc. Congestion control during communication with a private network
EP3410658B1 (fr) 2016-03-25 2020-06-24 Huawei Technologies Co., Ltd. Procédé et appareil de commande d'encombrement et dispositif correspondant
US10419968B2 (en) * 2016-03-30 2019-09-17 International Business Machines Corporation Dynamic selection of TCP congestion control for improved performances
US10097462B2 (en) * 2016-04-02 2018-10-09 Niciria, Inc. Throughput resilience during link failover
US11112784B2 (en) 2016-05-09 2021-09-07 Strong Force Iot Portfolio 2016, Llc Methods and systems for communications in an industrial internet of things data collection environment with large data sets
CN105827537B (zh) * 2016-06-01 2018-12-07 四川大学 一种基于quic协议的拥塞改进方法
US20170366466A1 (en) * 2016-06-20 2017-12-21 Mediatek Inc. Method of Reducing Transmission Control Protocol Acknowledgement and Wireless Device Using The Same
EP3476085A1 (fr) * 2016-06-22 2019-05-01 Martin Kuster Ordinateur de communication perfectionné
US10154431B2 (en) * 2016-09-27 2018-12-11 Verizon Patent And Licensing Inc. Congestion mitigation based on user device and base station condition information
US10341240B2 (en) * 2016-12-12 2019-07-02 Microsoft Technology Licensing, Llc Equation-based rate control using network delay for variable bitrate scenarios
US20180176136A1 (en) * 2016-12-19 2018-06-21 Mediatek Inc. TCP Bufferbloat Resolution
US11159834B2 (en) * 2016-12-21 2021-10-26 British Telecommunications Public Limited Company Managing congestion response during content delivery
CN110115042B (zh) 2016-12-29 2023-02-21 英国电讯有限公司 在网络中输送视频序列的方法、数据发送器
CN109218215B (zh) * 2017-06-29 2021-11-19 华为技术有限公司 一种报文传输的方法和网络设备
CN109309934B (zh) * 2017-07-27 2021-01-15 华为技术有限公司 一种拥塞控制方法及相关设备
CN109474444A (zh) * 2017-09-07 2019-03-15 中国移动通信有限公司研究院 一种传输控制协议tcp窗口的调整方法及调整设备
CN107749827B (zh) * 2017-10-30 2023-06-27 哈尔滨理工大学 基于网络状态分类的网络拥塞控制方法、装置及系统
CN108011780B (zh) * 2017-12-01 2019-01-22 北京百度网讯科技有限公司 一种数据传输速率测量方法、装置、设备和计算机可读介质
US10931587B2 (en) * 2017-12-08 2021-02-23 Reniac, Inc. Systems and methods for congestion control in a network
CN109905327B (zh) * 2017-12-11 2021-05-07 网宿科技股份有限公司 一种无线网络数据传输方法、发送端及接收端
CN108270691A (zh) * 2018-01-11 2018-07-10 电子科技大学 一种用于数据中心的可选择性显式拥塞通知标记方法
CN110247853A (zh) * 2018-03-09 2019-09-17 网宿科技股份有限公司 Tcp拥塞控制方法、系统、存储介质及网络服务器
TWI665890B (zh) * 2018-11-14 2019-07-11 中華電信股份有限公司 障礙偵測裝置以及障礙偵測方法
CN110120896B (zh) * 2019-04-15 2020-12-25 北京达佳互联信息技术有限公司 网络延迟控制方法、装置、电子设备及存储介质
CN111200843B (zh) * 2020-01-03 2023-09-22 青岛乾程科技股份有限公司 一种基于5g移动网络的跨层拥塞控制方法
CN111478893B (zh) * 2020-04-02 2022-06-28 中核武汉核电运行技术股份有限公司 一种慢速http攻击的检测方法
JP7452665B2 (ja) * 2020-08-14 2024-03-19 日本電信電話株式会社 通信システム、通信品質測定方法、ゲートウェイ装置及びプログラム
US11722412B1 (en) * 2020-09-28 2023-08-08 Amazon Technologies, Inc. Dynamically managing connection parameters among multiple computing devices
US11646957B1 (en) * 2020-12-04 2023-05-09 Amazon Technologies, Inc. Network packet loss period expansion
CN113010882B (zh) * 2021-03-18 2022-08-30 哈尔滨工业大学 一种适用于缓存丢失攻击的自定义位置顺序模式匹配方法
CN113364698B (zh) * 2021-06-08 2023-05-30 北京汇钧科技有限公司 带宽估计方法、系统、电子设备与存储介质
CN114039931B (zh) * 2021-10-27 2024-04-02 浙江大华技术股份有限公司 一种控制数据传输的方法、装置、设备及介质
CN114143271B (zh) * 2021-11-30 2024-04-02 百果园技术(新加坡)有限公司 一种基于拥塞检测的带宽估算方法及装置
CN114285800A (zh) * 2021-12-17 2022-04-05 瓴盛科技有限公司 一种tcp数据流的拥塞调整方法及装置
US11902404B1 (en) * 2022-06-10 2024-02-13 Juniper Networks, Inc. Retaining key parameters after a transmission control protocol (TCP) session flap
CN118233383B (zh) * 2024-05-22 2024-08-16 苏州元脑智能科技有限公司 拥塞控制方法及拥塞控制模型训练方法、交换机及介质

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20020071388A1 (en) * 2000-11-16 2002-06-13 Einar Bergsson Selectable network protocol
US20050237929A1 (en) * 2004-04-21 2005-10-27 National University Of Ireland Maynooth Congestion control in data networks

Family Cites Families (72)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6990069B1 (en) * 1997-02-24 2006-01-24 At&T Corp. System and method for improving transport protocol performance in communication networks having lossy links
US6097697A (en) 1998-07-17 2000-08-01 Sitara Networks, Inc. Congestion control
US7092355B1 (en) * 1999-06-30 2006-08-15 Network Physics, Inc. Method for controlling congested network flow
US6643259B1 (en) * 1999-11-12 2003-11-04 3Com Corporation Method for optimizing data transfer in a data network
US7200111B2 (en) 2000-08-31 2007-04-03 The Regents Of The University Of California Method for improving TCP performance over wireless links
US7130268B2 (en) 2000-10-17 2006-10-31 Saverio Mascolo End-to-end bandwidth estimation for congestion control in packet switching networks
US7099273B2 (en) 2001-04-12 2006-08-29 Bytemobile, Inc. Data transport acceleration and management within a network communication system
US7855966B2 (en) * 2001-07-16 2010-12-21 International Business Machines Corporation Network congestion detection and automatic fallback: methods, systems and program products
US7299280B2 (en) 2001-10-17 2007-11-20 The Regents Of University Of California Method and apparatus for TCP with faster recovery
US7630305B2 (en) * 2003-07-29 2009-12-08 Orbital Data Corporation TCP selective acknowledgements for communicating delivered and missed data packets
WO2004088858A2 (fr) 2003-03-29 2004-10-14 Regents Of University Of California Procede et dispositif permettant d'ameliorer la transmission de donnees
US7636321B1 (en) * 2003-05-12 2009-12-22 Sprint Communications Company L.P. Method and system for measuring round-trip time of packets in a communications network
US7436778B1 (en) * 2003-05-12 2008-10-14 Sprint Communications Company, L.P. Related-packet identification
JP3988682B2 (ja) * 2003-06-10 2007-10-10 ソニー株式会社 送信装置および方法、記録媒体、並びにプログラム
US7974195B2 (en) 2003-06-12 2011-07-05 California Institute Of Technology Method and apparatus for network congestion control
ITBA20030039A1 (it) 2003-08-29 2005-02-28 Grieco Luigi Alfredo Controllo di congestione rate-based del traffico entrante
US7925775B2 (en) 2004-04-07 2011-04-12 Sony Corporation TCP congestion control based on bandwidth estimation techniques
US8125910B2 (en) 2004-06-25 2012-02-28 Nec Corporation Communication system
US7656800B2 (en) * 2004-07-30 2010-02-02 Cisco Technology, Inc. Transmission control protocol (TCP)
WO2006023604A2 (fr) 2004-08-17 2006-03-02 California Institute Of Technology Procede et appareil de regulation de l'encombrement du reseau comprenant la gestion des files d'attente et des mesures du retard unidirectionnel
EP1821442A4 (fr) 2004-11-17 2011-07-20 Sharp Kk Appareil émetteur, appareil récepteur et système de communication
US7760638B2 (en) 2004-11-29 2010-07-20 Nec Corporation High-throughput communication system, communication terminal, session relay, and communication protocol
ATE413033T1 (de) * 2004-12-22 2008-11-15 Ericsson Telefon Ab L M Datenflusssteuerung mit doppelter quittierung
JP4147534B2 (ja) 2005-02-03 2008-09-10 日本電気株式会社 通信装置および通信方法
US7577097B2 (en) 2005-03-22 2009-08-18 Microsoft Corporation Compound transmission control protocol
KR100705445B1 (ko) * 2005-07-21 2007-04-10 한국전자통신연구원 무선 에드 혹 망에서의 자원 사용량 제어 방법 및 이를이용하는 통신 장치
US7778273B2 (en) * 2005-10-21 2010-08-17 Microsoft Corporation Application-level multicasting architecture
US7760633B2 (en) * 2005-11-30 2010-07-20 Cisco Technology, Inc. Transmission control protocol (TCP) congestion control using transmission delay components
EP1793553A1 (fr) * 2005-12-02 2007-06-06 Alcatel Lucent Un TCP hôte avec un module de convergence TCP
US8238242B2 (en) * 2006-02-27 2012-08-07 Telefonaktiebolaget Lm Ericsson (Publ) Flow control mechanism using local and global acknowledgements
CN101087244A (zh) * 2006-06-07 2007-12-12 华为技术有限公司 一种流控制传输中拥塞控制的实现方法
US8224255B2 (en) * 2006-08-22 2012-07-17 Embarq Holdings Company, Llc System and method for managing radio frequency windows
US7843831B2 (en) * 2006-08-22 2010-11-30 Embarq Holdings Company Llc System and method for routing data on a packet network
US8418016B2 (en) * 2006-10-05 2013-04-09 Ntt Docomo, Inc. Communication system, communication device, and communication method
WO2008047727A1 (fr) * 2006-10-17 2008-04-24 Panasonic Corporation Appareil de communication et procédé de communication
WO2008053841A1 (fr) * 2006-10-30 2008-05-08 Kyocera Corporation Appareil de communication, procédé de communication, appareil de contrôle de communication, appareil de communication sans fil, procédé de contrôle de communication et procédé de communication sans fil
JP5205819B2 (ja) * 2007-06-06 2013-06-05 日本電気株式会社 通信システムおよびそのパケットスケジューリング方法ならびに送信ノード
US7729249B2 (en) 2007-07-16 2010-06-01 Microsoft Corporation Systems and methods for improving TCP-friendliness of delay-based congestion control
EP2028798B1 (fr) * 2007-08-22 2012-05-02 Telefonaktiebolaget L M Ericsson (publ) Procédés et dispositifs de contrôle de transmission de données
JP5146725B2 (ja) 2007-09-19 2013-02-20 日本電気株式会社 通信装置および通信方法
US7839859B2 (en) * 2007-12-03 2010-11-23 Nec Laboratories America, Inc. Voice adaptive gateway pacing methods and systems for wireless multi-hop networks
EP2250771B1 (fr) * 2008-02-29 2019-04-24 Koninklijke Philips N.V. Optimisation de surveillance physiologique basée sur une qualité de signal disponible mais variable
US8837279B2 (en) * 2008-06-06 2014-09-16 Optis Wireless Technology, Llc Technique for improving congestion control
US8045463B2 (en) * 2008-07-30 2011-10-25 Microsoft Corporation Path estimation in a wireless mesh network
US8154996B2 (en) * 2008-09-11 2012-04-10 Juniper Networks, Inc. Methods and apparatus for flow control associated with multi-staged queues
US20100098047A1 (en) * 2008-10-21 2010-04-22 Tzero Technologies, Inc. Setting a data rate of encoded data of a transmitter
US20120147840A1 (en) * 2008-12-31 2012-06-14 Mediatek Inc. Method for boosting downlink transmission to mobile station and system utilizing the same
EP2219343A1 (fr) * 2009-02-12 2010-08-18 BRITISH TELECOMMUNICATIONS public limited company Modification de notification explicite de l'encobrement (ECN) par supprimer événements de encombrement expérimenté (CE)
US20120063493A1 (en) * 2009-03-06 2012-03-15 Yohei Hasegawa Transmission rate control method, transmission unit, and communication system
JP2010226455A (ja) * 2009-03-24 2010-10-07 Sony Corp ネットワーク通信装置
US8451862B2 (en) * 2009-07-15 2013-05-28 Qualcomm Incorporated Systems and methods for resource allocation serving communication requirements and fairness
US8509193B2 (en) * 2009-07-21 2013-08-13 Microsoft Corporation Packet aggregation
JP5402581B2 (ja) * 2009-12-01 2014-01-29 富士通株式会社 パケット中継装置および輻輳制御方法
US8625622B2 (en) * 2009-12-25 2014-01-07 Cisco Technology, Inc. Increasing transmission rate to a remote device in response to attributing information loss as not being a result of network congestion
CA2737107C (fr) 2010-04-13 2019-08-27 Jingyuan Wang Commande de congestion tcp pour reseaux heterogenes
US8514715B2 (en) 2010-04-22 2013-08-20 Akamai Technologies, Inc. Congestion window control based on queuing delay and packet loss
US9288137B2 (en) * 2010-05-09 2016-03-15 Citrix Systems, Inc. Systems and methods for allocation of classes of service to network connections corresponding to virtual channels
CN102006283B (zh) * 2010-10-21 2013-07-17 意法·爱立信半导体（北京）有限公司 数据传输的方法和装置
KR101173382B1 (ko) * 2010-10-29 2012-08-10 삼성에스디에스 주식회사 데이터 전송 방법 및 장치
US8824281B2 (en) * 2010-11-17 2014-09-02 At&T Intellectual Property I, L.P. Network-friendly transmission control protocol (TCP) methods, apparatus and articles of manufacture
US20120163167A1 (en) * 2010-12-27 2012-06-28 Symbol Technologies, Inc. Transmission control protocol optimization systems and methods for wireless networks
WO2012089110A1 (fr) * 2010-12-28 2012-07-05 The Chinese University Of Hong Kong Systèmes et procédés pour améliorer les performances d'un protocole tcp sur des réseaux ayant un produit bande passante-retard très élevé
CN102104908B (zh) * 2011-01-18 2014-05-07 华为技术有限公司 一种数据传输控制方法及设备
US8724471B2 (en) * 2011-03-02 2014-05-13 Mobidia Technology, Inc. Methods and systems for sliding bubble congestion control
US9178789B2 (en) 2011-08-19 2015-11-03 Futurewei Technologies, Inc. System and method for transmission control protocol slow-start
WO2013052649A1 (fr) * 2011-10-05 2013-04-11 Telcordia Technologies, Inc. Procédé et système d'attribution de bande passante par priorité, distribuée, dans des réseaux
US9042933B2 (en) * 2011-10-14 2015-05-26 Futurewei Technologies, Inc. Inter-cell interference coordination for wireless communication systems
US10292066B2 (en) * 2011-11-04 2019-05-14 Cisco Technology, Inc. System and method of modifying congestion control based on mobile system information
US20140164641A1 (en) * 2012-12-11 2014-06-12 The Hong Kong University Of Science And Technology Congestion control for data center traffic
JP6051832B2 (ja) * 2012-12-17 2016-12-27 富士通株式会社 通信装置，通信方法，及び通信プログラム
US20150236966A1 (en) * 2014-02-18 2015-08-20 Alcatel-Lucent Usa Inc. Control of congestion window size of an information transmission connection
JP5826892B1 (ja) * 2014-06-02 2015-12-02 ヤンマー株式会社 変化点検出装置、変化点検出方法、及びコンピュータプログラム

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20020071388A1 (en) * 2000-11-16 2002-06-13 Einar Bergsson Selectable network protocol
US20050237929A1 (en) * 2004-04-21 2005-10-27 National University Of Ireland Maynooth Congestion control in data networks

Non-Patent Citations (6)

* Cited by examiner, † Cited by third party
Title
IETF RFC 2018: TCP SELECTIVE ACKNOWLEDGMENT OPTIONS, October 1996 (1996-10-01)
IETF RFC 6582 ''THE NEWRENO MODIFICATION TO TCP'S FAST RECOVERY ALGORITHM,'', 2012
IETF RFC 6675: ''A CONSERVATIVE LOSS RECOVERY ALGORITHM BASED ON SELECTIVE ACKNOWLEDGMENT (SACK) FOR TCP, 2012
J. POSTEL, IETF RFC 793: TRANSMISSION CONTROL PROTOCOL,, 1981
K. NICHOLSV. JACOBSON: "Controlling queue delay", COMMUNICATIONS OF THE ACM, vol. 55, no. 7, 2012, pages 42 - 50, XP055115624, DOI: 10.1145/2209249.2209264
V. JACOBSON: "Congestion avoidance and control,", 1988, ACM SIGCOMM COMPUTER COMMUNICATION REVIEW

Also Published As

Publication number Publication date
EP3742687B1 (fr) 2024-09-11
CN106464598A (zh) 2017-02-22
US10516616B2 (en) 2019-12-24
US20240113973A1 (en) 2024-04-04
IL248275A0 (en) 2016-11-30
ES2666426T3 (es) 2018-05-04
DK3135009T3 (en) 2018-06-06
EP3135009A1 (fr) 2017-03-01
US11876714B2 (en) 2024-01-16
NO3135009T3 (fr) 2018-08-11
EP3319281B1 (fr) 2020-08-19
US20170195231A1 (en) 2017-07-06
EP3135009B1 (fr) 2018-03-14
EP3319281A1 (fr) 2018-05-09
US20220255863A1 (en) 2022-08-11
ES2845078T3 (es) 2021-07-23
IL248275B (en) 2020-07-30
US10263894B2 (en) 2019-04-16
WO2015161990A1 (fr) 2015-10-29
US11329920B2 (en) 2022-05-10
CN106464598B (zh) 2019-04-23
US20200092206A1 (en) 2020-03-19
US20190190831A1 (en) 2019-06-20

Similar Documents

Publication Publication Date Title
US11876714B2 (en) Method and apparatus for network congestion control based on transmission rate gradients
US7099273B2 (en) Data transport acceleration and management within a network communication system
Sinha et al. WTCP: A reliable transport protocol for wireless wide-area networks
US9112799B2 (en) Network packet loss processing method and apparatus
EP2387837B1 (fr) Maximum de l'utilisation de largeur de bande dans les réseaux avec latences élevées et les baisses de paquets utilisant le protocole tcp
EP1698087B1 (fr) Optimisation transparente pour un controle de flux tcp
US10064073B2 (en) Optimizing bandwidth of cognitive radios
US8416694B2 (en) Network feedback method and device
US20060209838A1 (en) Method and system for estimating average bandwidth in a communication network based on transmission control protocol
JP2006506866A (ja) データユニット送信機及びこの送信機の制御方法
US8565249B2 (en) Queue management system and methods
TWI308012B (en) Method for adaptive estimation of retransmission timeout in wireless communication systems
Kabir et al. Study of Different TCP Protocols in Wireless Network
Xiao TCP Congestion Control and its Variations
Pujeri et al. Survey of End-to-End TCP Congestion Control Protocols
Martignon Enhanced bandwidt estimation and loss differentiation in the TCP congestion control scheme.
Hong et al. Combining bandwidth estimate and explicit congestion notification for improving TCP over heterogeneous networks
Asplund et al. Partially Reliable Multimedia Transport

Legal Events

Date Code Title Description
PUAI Public reference made under article 153(3) epc to a published international application that has entered the european phase

Free format text: ORIGINAL CODE: 0009012

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: THE APPLICATION HAS BEEN PUBLISHED

AC Divisional application: reference to earlier application

Ref document number: 3135009

Country of ref document: EP

Kind code of ref document: P

Ref document number: 3319281

Country of ref document: EP

Kind code of ref document: P

AK Designated contracting states

Kind code of ref document: A1

Designated state(s): AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: REQUEST FOR EXAMINATION WAS MADE

17P Request for examination filed

Effective date: 20210525

RBV Designated contracting states (corrected)

Designated state(s): AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: EXAMINATION IS IN PROGRESS

17Q First examination report despatched

Effective date: 20211006

REG Reference to a national code

Ref document number: 602015089901

Country of ref document: DE

Ref country code: DE

Ref legal event code: R079

Free format text: PREVIOUS MAIN CLASS: H04L0012801000

Ipc: H04L0047110000

GRAP Despatch of communication of intention to grant a patent

Free format text: ORIGINAL CODE: EPIDOSNIGR1

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: GRANT OF PATENT IS INTENDED

RIC1 Information provided on ipc code assigned before grant

Ipc: H04L 47/11 20220101AFI20231025BHEP

INTG Intention to grant announced

Effective date: 20231123

GRAJ Information related to disapproval of communication of intention to grant by the applicant or resumption of examination proceedings by the epo deleted

Free format text: ORIGINAL CODE: EPIDOSDIGR1

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: EXAMINATION IS IN PROGRESS

GRAP Despatch of communication of intention to grant a patent

Free format text: ORIGINAL CODE: EPIDOSNIGR1

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: GRANT OF PATENT IS INTENDED

INTG Intention to grant announced

Effective date: 20240409

GRAS Grant fee paid

Free format text: ORIGINAL CODE: EPIDOSNIGR3

GRAA (expected) grant

Free format text: ORIGINAL CODE: 0009210

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: THE PATENT HAS BEEN GRANTED

AC Divisional application: reference to earlier application

Ref document number: 3135009

Country of ref document: EP

Kind code of ref document: P

Ref document number: 3319281

Country of ref document: EP

Kind code of ref document: P

AK Designated contracting states

Kind code of ref document: B1

Designated state(s): AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR

REG Reference to a national code

Ref country code: GB

Ref legal event code: FG4D

REG Reference to a national code

Ref country code: CH

Ref legal event code: EP

REG Reference to a national code

Ref country code: DE

Ref legal event code: R096

Ref document number: 602015089901

Country of ref document: DE

REG Reference to a national code

Ref country code: IE

Ref legal event code: FG4D